As documented in recent research, this review offers a thorough examination of the intricate subject of fingerprint authentication, including a wide range of issues and applications. Addressing problems like non-linear deformations and enhancing picture quality, which are frequently reduced by sophisticated improvement and alignment techniques are important components of fingerprint image authentication. Countering security concerns such as spoofing is a major focus of Automated Fingerprint Identification Systems and necessitates the use of sophisticated cryptographic techniques and liveness detection. In order to accomplish speedier identification processes, the paper emphasizes the advancements made in fingerprint indexing and retrieval, with a focus on deep learning technologies and minutiae-based methodologies. Furthermore, fingerprint authentication is used for a variety of age groups, including neonates, where it is essential for identification verification and the management of medical records. The paper also highlights the wider uses of fingerprint technology, such as improved crime detection skills, insights into age-related features, and contributions to medical diagnostics. This review provides a thorough overview of the latest developments and potential future directions in fingerprint authentication by combining state-of-the-art methodologies and analysing technical details, implementation challenges, and security issues. This captures the dynamic and important role of this biometric technology.

1. Introduction

In the rapidly evolving landscape of biometric technology, fingerprint authentication has emerged as a cornerstone of identity verification and security systems. The intricate and unique ridge patterns present on human fingertips provide a reliable and convenient means of confirming individuals’ identities. As societies transition towards dig- italization and interconnectedness, the role of fingerprint biometrics becomes increasingly crucial in ensuring secure access to systems, facilities, and personal information. This paper delves into the intricate web of fingerprint biometric research, offering a comprehensive overview of the chal- lenges faced, innovative solutions devised, and emerging trends that collectively shape the present and future of this dynamic field. Figure 1 depicts a few fingerprint images that have been collected through different technologies. Figure 1(a) represents fingerprint collected by ink and paper method, Figure 1(b) shows fingerprint collected by a digital scanner and Figure 1(c) shows fingerprint collected from a crime scene.

We will go through a number of fingerprint biometric dimensions in this exploration, each with its own opportunities and limitations. The challenge of authenticating unformatted fingerprint photos is significant because of non-linear deformations, different pressures during acquisition, and a range of environmental factors. [1]. In response, researchers have developed intricate algorithms harnessing machine learning, convolutional networks, and auto encoders to enhance image quality and improve matching accuracy. Security vulnerabilities inherent in Automated Fingerprint Identification Systems (AFIS) call for advanced strategies to safeguard against threats such as spoofing and data tampering. Biometric cryptosystems and cancellable templates are among the innovative solutions that fuse cryptographic techniques with biometric data, forging new frontiers in secure authentication [2]. The realm of fast fingerprint indexing and searching is another pivotal arena, demanding the fusion of speed and precision in matching large datasets. Leveraging fingerprint features and deep learning algorithms, researchers have crafted strategies to expedite the matching process while ensuring accurate results. Furthermore, the application of fingerprint biometrics extends beyond traditional identity verification. From diagnosing medical conditions through changes in fingerprint patterns to detecting drug consumption and even deter- mining handedness, the fingerprint’s unique attributes are being harnessed for a diverse array of purposes. As this exploration unfolds, it underscores not only the strides made in fingerprint biometrics but also the ethical considerations and societal implications that accompany the growing integration of biometric data into various aspects of modern life. As researchers continue to innovate and push the boundaries of what is possible, fingerprint biometrics remain a cornerstone in the pursuit of a secure and interconnected future [3].

The objective of this review is to comprehensively explore the landscape of fingerprint authentication as documented in previous literature. It aims to delve into the technical challenges encountered in fingerprint image authentication, including addressing non-linear deformations, enhancing image quality, and implementing alignment methods. The review also seeks to examine the security considerations within Automated Fingerprint Identification Systems (AFIS), focusing on strategies to combat spoofing through liveness detection and advanced cryptographic measures. Addition- ally, it aims to analyse the methodologies and advancements in rapid fingerprint indexing and retrieval, particularly em- phasizing minutiae-based approaches and the integration of deep learning techniques. Furthermore, the review intends to assess the diverse applications of fingerprint authentication, from identity verification in infants to its implications for medical history tracking, as well as its broader roles in medical diagnostics, age-related studies, handedness detection, and crime prevention. Through a thematic analysis of existing literature, this review aims to provide insights into the current state-of-the-art methodologies, implementation challenges, and security concerns within the field of fingerprint authentication.

To present a thorough overview of developments in the field of biometric authentication, we carefully reviewed publications published in the past ten years as part of the data collecting and analysis for this review study. Each study was methodically identified and categorized accord- ing to its purported benefits and drawbacks, accuracy measures, and procedures used. To comprehend these research’ methods to algorithmic processing, performance evalua- tion, and biometric data gathering, we carefully examined their techniques. The study aimed to evaluate the effi- cacy and accuracy rates of various biometric approaches, taking into account the advantages and disadvantages of each method. Through the consolidation of this data, we were able to identify patterns and breakthroughs, evaluate the development of biometric technologies, and provide perspectives on the condition of the technology sector at large. This comprehensive study seeks to offer a fair assessment of the advancements made, point out areas in need of development, and determine future paths for biometric authentication research.

In this study, we attempt to investigate various state-of-the-art developments in finger print biometric practical application. The format of this essay is as follows: The difficulties and security risks associated with latent fingerprint authentication and automatic fingerprint authentication systems are discussed in Section 2. The method for all-purpose fingerprint user identification is described in Section 3. Fast fingerprint indexing and searching are covered in Section 4. The difficulties and solutions for analysing and identifying infants’ fingerprints are described in Section 5. Multi-modal biometric techniques are discussed in Section 6. Comparative comparison of various approaches is the main topic of Section 7. Section 8 concludes with some final thoughts.

2. Challenges in Authentication

2.1. Authentication of Unformatted Fingerprint Image

In the field of fingerprint authentication, raw fingerprint images can be obtained from different sources, such as rolled ink prints, live scans, and latent prints. These images often suffer from non-linear deformations and poor quality, mak- ing authentication challenging. Non-linear deformations can arise due to variations in finger pressure and improper finger placement during image acquisition. These deformations can lead to unsatisfactory matching scores and impact the accuracy of fingerprint authentication systems. Low-quality fingerprint images can also result from factors like image scanner noise, partial prints collected from crime scenes (latent prints), and various skin conditions. These low-quality prints can be categorized as dry, wet, damaged, dotted, and blurred. Dry prints have minimal contact with the scanning surface, while wet prints result in smudged ridges and valleys due to extensive contact. Damaged prints may arise from scars or skin issues, while dotted prints are caused by excessive sweating. Blurred prints occur when motion or unclear ridges during scanning lead to image blurriness. To address these challenges, various techniques have been developed for fingerprint enhancement and alignment. Traditional methods such as Gabor filtering and adaptive boosted spectral filtering are commonly used for enhancing fingerprint images. Alignment techniques and cluster-based methods aim to rectify non-linear deforma- tions, improving the accuracy of matching in automated fingerprint identification systems (AFIS). Figure 2 shows fingerprint feature extraction using convolution neural network, where three parallel CNN extract features from the fingerprint image and after the merging of the three features to form a unique feature descriptor, which is passed to a classifier for classification task.

Because of their high dimensionality, minutiae descriptors, which are widely used in fingerprint identification algorithms, face considerable hurdles. These descriptors are robust for fingerprint pattern identification because they capture fine features. However, the noise and volatility present in real-world settings might have a significant impact on their performance. Noise may cause mismatches during matching procedures by distorting minute points. This noise can originate from various sources, such as defective sensors or low-quality images. Accurate identification is made more difficult by unpredictable variations in finger- print traits, such as ageing or changes brought on by injuries. Thus, even though minutiae descriptors are excellent at collecting distinctive fingerprint features, their effectiveness depends on how well noise and unpredictability are handled to provide safe and dependable biometric identification systems.

A novel latent fingerprint matching approach is presented by [4], which makes use of a descriptor-based Hough transform for latent prints together with a robust alignment technique. Their approach achieved a noteworthy rank-1 accuracy of 53.5% on NIST SD27, outperforming two commercial matchers and a non-commercial algorithm. Singular point-based alignment is a problem, as performance depends on the quantity of minutiae and the overall print quality. Refining features with top-down information from an exemplar print, in [5], the authors presented a feedback system for latent fingerprint matching. When combined with a cutting-edge matcher, it improved identification accuracy in the NIST SD27 and WVU databases by 0.5–3.5%. Finding a balance between the advantages of feedback and the complexity of the system, as well as the example print quality, is a challenge. Deformable Minutiae Clustering was proposed by [6] for latent fingerprint recognition, improving minutiae matching via cluster merging and Thin Plate Spline modelling. It outperformed previous techniques with up to 85.6% (Cylinder-Codes) and 83.3% (m-triplets) accuracy across different fingerprint databases. Scalability issues for latent-to-latent identification and algorithm speed are among the limitations. Using feature selection and random decision forest classification, the authors in [7] provided an adaptive latent fingerprint segmentation approach that achieved state-of-the-art performance on three databases. In order to balance accuracy and selection time, their method includes a unique SIVV-based metric for segmentation evaluation and modified RELIEF feature selection. Constraints about the applicability of the SIVV metric and assumptions about consistent ground truth are examples of limitations. A collaborative filtering paradigm for fingerprint enhancement was proposed by [8]. It involves pre-enhancing using Gabor filters and spatial patch-based enhancement utilizing spectral diffusion after that. Test findings on FVC2004 datasets showed that this approach was better than other approaches such as the Gabor filter and VeriFinger Algorithm Demo. The study is limited by fixed patch size effects and sensitivity to input fingerprint quality, which calls for additional investigation to fully evaluate resilience. FingerNet, a CNN-based method for latent fingerprint enhancement, was introduced by [9]. It has shared convolution and deconvolution layers for orientation guiding and noise reduction. It demonstrated potential for improvement with larger datasets and ground truth ROI annotations, as demonstrated by its top-1 matching accuracy of 47.7% when tested on NIST SD27 using data collected from SD4. Using Convolutional Neural Networks (ConvNets) for ridge flow estimation and minutia descriptor extraction. In [10], the authors proposed an automated latent finger- print recognition system that achieved rank 1 identification accuracies of 64.7% and 75.3% on NIST SD27 and WVU latent databases. Poor ridge quality, background noise, a tiny friction ridge area, reliance on manual ROI selection, long processing times, and scaling issues are some of the limitations. A technique to improve Automated Fingerprint Identification Systems (AFIS) by utilizing uncommon minutiae was suggested by [11]. This involves altering similarity scores that are based on least squares fitting mistakes in order to improve matching accuracy. It greatly improves the rank identification accuracy of minutiae-based matchers, as demonstrated by tests conducted on the GCDB forensic database. Future study on generalization and database size effects is suggested by the challenges of dataset restrictions and manual intervention for latent AFIS. An end-to-end latent fingerprint recognition system with automated ROI cropping, minutiae extraction based on deep learning, and texture template creation was promoted by [12]. Tested against a background of rolled prints on many datasets, yielding rank-1 retrieval rates varying from 7.6% to 69.4%. Among the drawbacks are difficulties in identifying fine details on low-quality photos, trimming dry latents, and the requirement for more varied operating databases for thorough training. By utilizing fine-coarse parallelism and asynchronous processing for latent fingerprint recognition. In [13], the authors invented ALFI and achieved a 22x speed gain over state-of-the-art approaches with equivalent ac- curacy. Tested on NIST SD27, they report accuracy using Equal Error Rate and F1-score, recognizing dataset size constraints, restricted classification strategies, and unknown uses beyond fingerprints, such as DNA analysis. LQMetric, an automated method for evaluating latent fingerprint quality and forecasting the probability of a rank-1 hit in the FBI’s NGI AFIS system, was suggested by [14]. It produced a 61.4% match with human examiner assessments using local clarity maps and image analysis tools; it worked well for NGI searches but was not uniformly applicable to other AFIS methods. For improving latent fingerprint images. In [15], the authors provide a progressive GAN-based technique that consists of off-line training and repeated on-line testing phases. Based on CMC curve metrics, evaluation on the NIST SD27 dataset demonstrates better performance than previous models. Constraints on dataset quantity, processing cost, and possible inefficiency with subtle ridge features in latent prints are some of the limitations. An automated latent fingerprint identification system using DCNN-FFT augmentation for minutiae extraction and matching was described by [16]. It achieved 100% rank-1 identification on the FVC2002 and FVC2004 databases and 84.5% on the NIST SD27. The results show increased recall, F1 scores, and precision; however, computational time and dataset diversity are acknowledged as constraints. Using synthetic data for training. In [17], the author provide a deep nested UNets architecture for automated latent fingerprint segmentation and improvement. It performs better than current methods in fingerprint recognition and segmentation accuracy when tested on the NIST SD27 and IIITD-MOLF databases. One of the main drawbacks is the lack of high- quality and low-quality coupled fingerprint picture pairs in the current databases. This can be overcome by creating synthetic latent images for training. Using dense fingerprint patch alignment and matching. In [18], the authors promoted a non-minutia latent fingerprint registration method that outperformed previous techniques in registration when tested on the NIST27 and MOLF datasets. Its drawbacks include its dependence on 2D data, higher processing requirements, and possible inadequacy for poor-quality picture data. Using an algorithm based on Lindeberg’s scale selection method. In [19], the authors revealed integrating pores (level 3 characteristics) alongside minutiae for latent fingerprint identification. When combined with pores, as opposed to just minutiae matching, greatly improves recognition rates on the IIITD Latent database. A small dataset size, a focus plane restriction for pore extraction, and computational costs that affect matching time are among the limitations. For latent fingerprint identification. In [20], the authors proposed the “Ratio of Minutiae Triangles” (RMT) algorithm, which makes use of local minutiae arrangements without pre-alignment. tested with Rank-1 recognition accuracy of 80% and 63.8% on the FVC2004 and NIST SD27 databases, respectively. Poor quality latent cases and the requirement for better feature extraction and matching phase integration for large databases are acknowledged constraints. In order to achieve accurate latent fingerprint enhancement and segmentation. In [21], the researchers proposed a hybrid model that combines the Chan-Vese approach for segmentation and Edge Adaptive Directional Total Variation (EDTV) for enhancement. Accurate Rank-1 identification was achieved with 61.24% and 70.16% on NIST SD27 and WVU DB databases, respectively. Restrictions include the size of the database and the requirement for additional testing on intricate latent fingerprints. Using Chan-Vese, an adaptive latent fingerprint segmentation and matching strategy based on the EDTV model was given by [22]. On the NIST SD27 dataset, this approach achieved AUCs of 72% and 70.59% for ROC and CMC curves, respectively. Limitation includes poor accuracy of existing techniques for latent fingerprint segmentation. Using scale and rotation invariant minutiae characteristics. In [23], the authors proposed an automatic latent fingerprint identifica- tion system that significantly improved Rank-1 identification accuracy on NIST SD27 and FVC2004 datasets. On NIST SD27, CLMP-NRS produced the maximum Rank-1 accuracy of 93.80%. Handling incomplete fingerprints with limited minutiae characteristics and relying on the quantity of retrieved minutiae for matching performance are two limitations. MinNet, a minutia patch embedding network for latent fingerprint identification that optimizes spatial and angular minutiae distribution, is introduced by [24]. Reaching cutting-edge outcomes, assessed on many public and private datasets, such as FVC-Latent and NIST SD27, with rank-1 accuracies of 85.88% and 92.39%, respectively. Its limitations include the inability to handle fingerprints that are partially or substantially deformed and its reliance on high-quality latent pictures for precise minutiae extraction. To achieve state-of-the-art accuracy and efficiency in latent fingerprint indexing, the authors in [25] offer a multi-scale fixed-length representation technique. performed better than other indexing techniques like PDC and DeepPrint when tested on a variety of datasets, including Hisign and NIST SD27. possibility performance variance on simulated fingerprints and scope-limited optimization are among the limitations, indicating possibility for more improvements in feature representation. Using a residual encoder-decoder architecture and frequency-domain loss function optimization, the authors in [26] proposed a deep learning model for latent fingerprint enhancement. superior than current methods in terms of rank-25 and rank-50 accuracy when evaluated using the IIIT-Delhi MOLF database. The ridge- based approach of the technology and the requirement for additional refinement in deleting undesired portions of images and testing on a variety of databases are its limitations. With an emphasis on fairness in prediction and decision- making, the authors in [27] provided a technique to and reduce biases in automated algorithms working with latent fingerprint pictures. They evaluate the effect on automatic matching of latent fingerprints using covariate-specific ROC curves produced from regression models taking quality and demographics into account. Results demonstrate that, compared to assessments without quality, the suggested covariate-adjusted ROC curves conditioned on image qual- ity and demographics offer a more informative assessment. The quality measurement algorithm’s underlying assump- tions and dataset restrictions are examples of limitations. It is believed that more research on a variety of datasets is crucial for method validation. ULPrint, a Universal Latent Fingerprint Enhancer, was introduced by [28], who used Mix Visual Transformer (MiT) SegFormer-B5 encoder ar- chitecture and Ridge Segmentation with UNet. By using directed blending of predicted ridge masks, the technique improves latent fingerprints and addresses issues with a variety of latent types. Tests conducted on both synthetic and actual datasets show notable gains in accuracy; nonethe- less, there are still issues such as annotator subjectivity and limited databases. FingerGAN, a GAN-based technique for latent fingerprint enhancement, was presented by [29]. It optimizes minute information through adversarial, perceptual, and reconstruction losses. Tested on NIST SD27 and IIIT-Delhi MOLF, FingerGAN outperforms state-of-the-art techniques, however it has drawbacks such as computational complexity and dependence on high-quality rolled fingerprints for data production. For latent fingerprint recognition, the authors in [30] suggest a hybrid method integrating local (minutiae and virtual minutiae) and global features (AFR-Net embeddings). With a multi-stage matching paradigm and Squeeze U-Net CNN for augmentation, they attain an average rank-1 retrieval rate of 71.22% on multiple datasets. Failure instances including severe rotations and overlapping patterns are noted, as are challenges such as low contrast and occlusion. ACSACO, a hybrid approach combining Cuckoo Search and Ant Colony Optimization for latent fingerprint identification, was proposed by [31]. AC- SACO beats individual algorithms when tested on the NIST SD-27 dataset, achieving excellent precision and recall for prints of good quality but lower accuracy for prints of poor or ugly quality. Complex backdrop problems and overlapping prints are among the limitations, indicating the need for additional optimization strategies and wider dataset validation. In order to enhance DeepPrint for recognition systems, in [32], the authors developed a CycleGAN-based technique to create artificial latent fingerprints. Their approach improves TDR at 0.01% FAR from 23.64% (MSP) to 45.45% (synthetic) when tested on NIST SD27. Reliance on a single matcher and the lack of functional latent databases are among the limitations.

Latent fingerprint identification is the painstaking process of locating and examining fingerprints from crime scenes, frequently with the aid of cutting-edge methods to improve their contrast and visibility. New levels of security are introduced when these latent prints are scanned and input into automatic fingerprint identification systems (AFIS). This meticulous manual process is essential for correct identification. Although AFIS significantly improves fingerprint matchings scalability and efficiency, it also brings with it potential vulnerabilities that were not as noticeable in manual systems. Problems like algorithmic flaws, spoofing attempts, and data breaches emphasize the necessity of strict security protocols to preserve the accuracy of fingerprint data. In order to guarantee that the improvements in fingerprint identification technology do not jeopardize its dependability and efficacy, it is crucial to comprehend the shift from manual latent fingerprint analysis to automated methods.

2.2. Security Threats in AFIS

Digital user authentication through fingerprint biometrics relies on Automated Fingerprint Identification Systems (AFIS) to verify individuals’ identities. The concept of AFIS was initially developed by the US Federal Bureau of Investigation (FBI). The AFIS process can be divided into two phases: Enrolment and Identification. In the Enrolment phase, users’ fingerprints are registered in the system. The Identification phase matches query fingerprints against stored templates for authentication. During Enrolment, fingerprint images undergo preprocessing, feature extraction, and template creation. In the Identification phase, query fingerprints are processed and matched against stored templates to identify valid users. Securing a biometric system is crucial due to the non-changeable nature of biometric traits. An AFIS needs robust security measures to safeguard user information from potential attackers. Various components within an AFIS can be exploited by attackers. These attack points include potential vulnerabilities in the biometric data acquisition process, template extraction, and communication channels. Several known attacks can compromise the security of an AFIS, including spoofing (presenting fake biometric data), exploiting similarity, zero-effort attempts (using attacker’s own biometric to impersonate a legitimate user), physical destruction of the biometric sensor, replay attacks (intercepting and replaying biometric signals), communication channel attacks (cutting or altering channels), and more. These attacks can lead to unauthorized access, denial of service, and other security breaches.

2.2.1. Liveness Detection

To address these security challenges, researchers have developed preventive measures and countermeasures, such as liveness detection to determine if a fingerprint is from a living person or a spoof. Various methods have been pro- posed to detect spoof attacks that use materials like wood, glue, and gelatine to fabricate fingerprint spoofs. Figure 3 shows the process of liveness detection using auto encoder where live and spoof fingerprint images are passed to a encoder to learnt two different latent representation of the image and then it is passed to decoder to reconstruct the live and spoof fingerprint, this way the architecture knows the difference between a spoof and live image and can detect them.

In order to improve the resilience of biometric authentication systems against spoofing attacks, the authors in [33] present an approach that combines SURF, PHOG, and Gabor wavelets with low-level characteristics and shape analysis to detect liveness in fingerprint images. Achieving an average EER of 3.95%, their algorithm, which uses PCA for dimensionality reduction and various classifiers, such as SVM and Random Trees—beams above the previous record of 9.625%, validated across several databases, including LivDet 2011 and LivDet 2013. However, more testing on a range of sensors and materials is acknowledged for a thorough robustness assessment. A software-based fingerprint liveness detection technique is presented by [34]. It uses image gradient co-occurrence arrays for feature extraction and SVM classification. It attains greater accuracy on LivDet09DB and LivDet11DB datasets than state-of-the-art methods, in spite of drawbacks like quantization-induced information loss and high-dimensional feature vectors derived from higher-order co-occurrence arrays. Convolutional neural networks could improve gradient measurement; however, dimensionality reduction methods could be needed to handle higher-order arrays. In order to achieve state-of-the-art accuracy across LivDet datasets, the authors in [35] proposed a CNN-based technique for fingerprint spoof attack detection that makes use of local patches centred on minutiae for fine-grained analysis. The approach shows effectiveness against different spoof materials and testing settings, with an average accuracy of 99.03% on LivDet 2015, beating the 95.51% of competition winners. The need for more diverse datasets and ethical concerns about the use of biometric data are two acknowledged constraints, though. A fingerprint liveness detection technique using a BP neural network and difference co-occurrence matrices for improved textural features was suggested by [36]. The system builds input data based on these matrices and uses pre-trained networks to forecast classification accuracy, achieving higher detection accuracy on the LivDet 2013 database over earlier techniques. One of the limitations is the lack of diagonal direction difference co-occurrence matrices, which could more accurately represent the properties of the image. For real-time fingerprint liveness detection, the authors in [37] promoted a semi-supervised stacked autoencoder-based method that substitutes learnt representations for hand-designed features. Exhibiting efficaciousness on the LivDet 2011 and 2013 datasets, the approach attains satisfactory performance, but with a recognition of constraints including dataset magnitude, model initialization, and difficulties with substandard quality, excessive brightness, or noisy images. Adversarial attacks on deep neural network-based fingerprint liveness detection were described by [38], ex- posing flaws in cutting-edge models. By applying FGSM, MI-FGSM, and Deepfool techniques, they illustrated how models might mistakenly identify phony fingerprints as real ones with minute variations. The evaluation of the LivDet 2013 and LivDet 2015 datasets made clear that deep learning applications require more reliable detection models and anti- adversarial techniques. With a score-level fusion approach that combines liveness detection and fingerprint matching, the authors in [39] took first place with 96.88% accuracy in the Fingerprint Liveness Detection Competition 2019. It was evaluated using the LivDet2015 and LivDet2019 datasets, taking into account restrictions on finger pressure and device time that may effect detection accuracy. It used ONNS for similarity, Slim-ResCNN for FLD score, and LR classifiers for fusion. A DenseNet optimized with a genetic algorithm was provided by [40] for fingerprint liveness detection, and it achieved 98.22% accuracy on a mixed Livdet dataset. De- spite constraints in dataset size and computational cost for real-time applications, the method outperforms state-of-the- art efforts on LivDet 2009, 2011, 2013, and 2015 datasets by utilizing ROI extraction and specialized mutation operators. A liveness identification approach utilizing the Circular Gabor Wavelet (CGW) algorithm and SVM was presented by [41], which achieved 99.968% accuracy in differentiating between real and false fingerprints. The approach, which has been tested on 272 samples from optical and capacitive sensors, yields encouraging results but still needs to be assessed further against spoofing techniques used in the real world. A multi-modal liveness detection method incorporating fingerprint and iris inputs, leveraging statistical texture features and spatial analysis, was advocated by [42]. The approach outperforms sum-rule and product- based approaches with high precision (94.7%) for fingerprint detection and 97.8% accuracy for decision-level fusion. recognizing its limits, such as its reliance on dataset size and its usefulness against particular attack types. Using the LivDet 2013 and 2015 datasets, the authors in [43] presented FLDNet, a lightweight CNN architecture for fingerprint liveness detection that achieves state-of-the-art performance. FLDNet solves problems such as accuracy on tiny size fin- gerprints and robustness against unknown spoof materials, and achieves 1.76% Average Classification Error (ACE) over all sensors with a redesigned dense block and attention pooling layer. A one-class convolutional auto encoder for fingerprint presentation attack detection is proposed by [44], and it achieves a D-EER of 2.00% on a dataset of 24,050 fingerprint images. Among the method’s drawbacks include its inability to generalize to additional modalities, the requirement for distinct models for various attack substrates, and uncertainties about robustness and adversarial attacks. A weighted multi-modal CNN-based FLD technique was introduced by [45], which used ROI operation and feature fusion for improved performance. It achieves excellent accuracy in different evaluation circumstances, outperforming current approaches on LivDet 2011, 2013, 2015, and NUAA datasets. Two drawbacks are the need for extensive training datasets and performance variability brought on by sensor diversity. EaZy learning, an adaptive ensemble learning model for fingerprint liveness detection, was suggested by [46]. On LivDet 2011, 2013, and 2015 datasets, it achieved average accuracies of 60.49% and 67.80%. It solves drawbacks such as reliance on clustering techniques and dataset size, as well as potential performance concerns in unknown contexts due to lack of variety in training data by clustering training data and integrating predictions using weighted majority voting. A novel approach using transformers and GANs to improve fingerprint presentation attack detection (PAD) generalization across sensors and materials was developed by [47]. With LivDet2015, the approach achieves an accuracy gain from 68.52% to 83.12%, addressing the limits of accuracy degradation in cross-sensor situations and inadequate generalization. Using a “transient liveness factor” (TLF) and picture quality measures, the authors in [48] suggest a person-specific FPAD technique that achieves 100% accuracy in spoof presentation detection. They admit constraints such as the short dataset size and the difficulty of generalizing across different materials while conducting experiments with a dataset of 30 live photos and 138 spoof samples from various attackers. In comparison to previous methods, the authors in [49] proposed a Fingerprint Liveness Detection (FLD) technique that improved accuracy by 1.0% on average by integrating AlexNet, VGG16, and ResNet CNNs with a genetic algorithm for feature weighting. Enhancing detection performance on several Livedet datasets, the strategy addresses the drawbacks of fixed-scale inputs and possible challenges in training the model due to over-abundance of features. A CNN-based fingerprint liveness detection technique was suggested by [50], which outperformed SVM and CNN+SVM hybrid techniques on the LivDet2015 dataset. Measured by precision, specificity, F1 score, and accuracy metrics, preprocessing approaches and feature extraction methods improve classification accuracy. Given that different materials provide different obstacles when it comes to creating false fingerprints, more study is necessary to improve classification performance in these scenarios. A multi-filter framework for finger- print liveness detection utilizing hand-crafted features was promoted by [51], who achieved improved performance on LivDet test datasets. The method achieves an average accuracy of 99.15% and an average classification error of 0.85% on the LivDet 2015 dataset, outperforming state-of- the-art techniques. It involves procedures such as data augmentation, preprocessing, feature fusion, and dimensionality reduction. The authors recognize that lengthy processing times and complex parameter setting are real usability barriers. In order to counteract forced and phony fingerprint attacks, the authors in [52] introduced MFAS, a Micro-Behavioural Fingerprint Analysis System that records fingertip micro-behaviour throughout time. After being tested on 24 subjects, MFAS achieves 100% accuracy in identifying voluntary versus forced fingerprint placements, as well as 100% true positives. Though encouraging, the system can run into problems in realistic situations, requiring more study to improve technology and comprehend its application. The incremental learning model A-iLearn, pro- posed by [53], addresses the stability-plasticity conundrum in spoof fingerprint detection by adaptively integrating base classifiers. A-iLearn increases accuracy on new fake materials by up to 49.57% without appreciable stability loss when compared to baseline models. It is tested on LivDet 2011, 2013, and 2015 datasets using a variety of spoof materials and sensors. Some drawbacks include the possibility of over fitting and the requirement for more research into feature integration techniques. By combining texture and sweating pore characteristics, the authors in [54] provided a static-based fingerprint liveness detection technique. An auto encoder minimizes feature dimensionality for binary classification using a softmax classifier by quantifying textural properties and pore activity. Testing the approach on LivDet 2013 and LivDet 2015, with an ACE of 0.11-0.24%, it recognizes the difficulties in maintaining several algorithms for hardware-based detection while addressing restricted pore feature-based techniques. A lightweight fingerprint liveness detection network called LFLDNet is proposed by [55]. It makes use of style transfer, foreground extraction, and an enhanced ResNet with MHSA. 95.27% accuracy on small-area fingerprints and 1.72% average classification error across sensors were achieved during evaluation on the LivDet2011, LivDet2013, and LivDet2015 datasets. The authors intend to investigate improved FLD technologies and GAN-based models for cross-sensor generalization against fingerprint deception assaults, while acknowledging the possibility of speed variations. Using CNNs and adversarial data augmentation, the authors in [56] proposed an FPAD technique and won the LivDet2021 competition. The method uses clean and adversarial fingerprints for a three- stage training process that yields good accuracy metrics and an EER of 0.036. The size restrictions of the dataset and possible susceptibilities to adversarial assaults are acknowledged by the authors. Using LPDJH image descriptors and deep learning, the authors in [57] introduced a fingerprint liveness detection technique that achieved good accuracy across LivDet datasets. Their approach shows competitive results on the LivDet 2011 dataset, with an average EER of 3.95%. Because there aren’t enough test samples, the authors point out possible difficulties in identifying low-resolution fingerprints and calculating accuracy for particular sensors. A CNN-based fingerprint liveness detection technique was introduced by [58] using the Socofing dataset. The model performed well, achieving 98.964% accuracy with a FAR of 0.215% and a FRR of 7.251%. Computational limitations, inconsistent fingerprint quality, and unequal dataset distribution are some of the restrictions.

By addressing the crucial problem of separating real biometric samples from fake or artificial copies, fingerprint liveness detection makes sure that the system authenticates the submitted fingerprints. By stopping unauthorized access, this procedure is crucial for preserving the integrity of bio- metric systems. However, biometric cryptographic systems use biometric information like fingerprints to strengthen security by using cryptographic methods. This way, bio- metric identifiers are safely encrypted and shielded from unwanted access. The interdependence of liveness detection technologies and the efficacy of biometric cryptography systems highlights the relationship between these two domains. Robust liveness detection essentially acts as a cornerstone supporting the security of biometric cryptography systems, emphasizing the need for integrated solutions that handle biometric sample authenticity as well as biometric data management securely.

2.2.2. Biometric Cryptosystem

Biometric cryptosystems are mechanisms that enhance the security of biometric systems by embedding biometric templates with secret keys or auxiliary data. This approach ensures protection against malicious use and data tampering. Figure 4 shows the block diagram of biometric cryptosystem, which consist of two parts one is for enrolment where template are generated from the fingerprint images and stored in the dataset and the second part is authentication where template is generated from a query fingerprint and matched with the stored template to provide access to a system.

Pair-polar minutiae structures were used by [59] to present an alignment-free fingerprint cryptosystem. Their approach achieves robust security and improves privacy by changing minute structures. It has been tested on several databases and performs better than other systems in terms of Genuine Acceptance Rate (GAR) and False Acceptance Rate (FAR); nonetheless, a security risk arises from its dependence on the complexity of brute force attacks. In order to maximize security and efficiency, the authors in [60] suggest a feature-level sequential fusion algorithm for biometric cryptosystems. Experiments conducted on a finger vein database show that the algorithm performs better than the OR rule, decreasing input needs and obtaining a 1.47% False Acceptance Rate. Although the paper admits its limits in analysing some external security concerns, it shows that it is resilient to well-considered attacks. An ECC-free biometric key binding approach employing graph- based Hamming Embedding (GHE) and minutia-vicinity decomposition (MVD) for fingerprint minutiae-based representation was proposed by [61]. The strategy obtains GAR of 89% (FRR=11%, FAR=0.16%) for DB1 and GAR of 97% (FRR=3%, FAR=0.061%) for DB2, when tested on the FVC2002 datasets. The method balances security and performance trade-offs, but it is still susceptible to privacy attacks such as ARM and SKI, and it can only match fingerprint photos of the same finger taken under various circumstances or at different times. In [62], the authors provide a method to improve biometric template security that is based on a 2D logistic sine map (2DLSM). The technique employs diffusion and confusion processes, demonstrating effectiveness against a range of threats, by using chaotic streams from 2DLSM. Effectiveness has been evaluated on iris and finger- print templates; nevertheless, real-time template production problems and noise sensitivity are among the constraints. A safe cryptographic authentication approach utilizing bio- metrics and the discrete logarithm problem was suggested by [63]. The approach produces cryptographic keys using a function F using iris codes, which allows it to handle mistake bits and achieve high accuracy rates. Promising results are shown by the CASIA iris database evaluation, which also notes that iris codes still require work in order to be improved in the future. BioKEY, a biometric-based cryptographic key generation technique based on convolution coding principles, was promoted by [64]. Unique cryptographic keys are created without saving templates by using fingerprint characteristics and applying convolution coding to specific locations. Tests conducted on common fingerprint databases such as MIAS, FVC2002, and FVC2004 show good accuracy and effectiveness; nevertheless, there are certain drawbacks such as difficulties with noisy or hazy fingerprints and a higher processing overhead in comparison to other techniques. ECC-based mutual authentication strategy for Smart Grid communications with biometric authentication integration for improved security and privacy was proposed by [65]. To safeguard against different types of security risks, the protocol consists of start-up, registration, and authentication stages. Notwithstanding its shortcomings, which include parameter sensitivity and vulnerability to certain assaults, the system shows efficiency through communication and computing cost comparisons with current protocols even if it does not offer precise accuracy measurements. In order to safeguard templates in multi-modal biometric systems, the authors in [66] provide a fuzzy vault technique that uses encryption to keep authentic templates safe from copycats. The approach satisfies optimal biometric security requirements with low FAR (0.1062) and zero FRR when tested on a simulated face and fingerprint database. In order to get the best security and accuracy, the authors recommend avoiding function creep attacks and balancing key length with chaff points. Hyper elliptic curve cryptography (HECC) for template encryption was used by [67] to introduce an improved iris recognition technique that achieves great security and accuracy. With the use of fuzzy logic matching and 2D Gabor filter for feature extraction, the system achieves 99.74% maximum accuracy and short identification time. Superior performance with reduced FAR, FRR, and EER measures is demonstrated by evaluation on the IITD and CASIA Iris V-4 iris datasets. A multi biometric cryptosystem for client-server network authentication is presented by [68], guaranteeing computation security and privacy. The approach enhances efficiency by achieving equivalent accuracy with less user input by using iris and fingerprint modalities from a dataset of 100 participants. The authors request protocol enhancements for increased security and verification speed, acknowledging the limitations in accuracy caused by biometric variability. A multi-biometric template security technique based on graph creation was proposed by [69] for cloud authentication. A branching factor graph with a low Equal Error Rate (ERR) of 0.66% is created using fingerprint and palm print features. The authors point out important benefits over conventional authentication techniques while also acknowledging limitations with regard to the encryption process and the quality of the input sample. Using the Diffie-Hellman algorithm and minute characteristics, the authors in [70] proposed a fingerprint-based crypto-biometric system for secure communication. Tested using FVC2002 and the NIST special database 4, the system reports metrics such as FAR, FRR, GAR, and EER, demonstrating privacy protection. The authors have acknowledged the restrictions pertaining to the irreversibility of biometric data, possible data distortions, and key creation. A multi-modal biometric cryptosystem utilizing fingerprint and ear characteristics was promoted by [71], which included preprocessing, feature extraction, and classification stages. The approach achieves excellent accuracy (98.76%) with a dataset that includes photos of ears and fingerprints. Performance is analysed using metrics such as sensitivity, specificity, accuracy, false positive rate, and false negative rate. The lack of multi-modal databases and the dependence on multi-modal systems because particular bio- metrics, such as ear characteristics, are unreliable are among the limitations. However, the method offers an efficient fix for identity and security systems. A biometric cryptosystem using random projection and back propagation neural net- work (BPNN) for template protection is introduced by [72]. Original biometric traits are transformed into unlinkable projected vectors by random projection, and BPNN is then used to map these vectors to secure keys. Better security and performance than current techniques are demonstrated via experimental assessment on a variety of biometric datasets. The authors stress the necessity for more investigation into novel ciphers for biometric template security and the improvement of security through multiple-biometric template protection. A novel fingerprint biometric cryptosystem using fuzzy commitment and CNN-based automated texture descriptor discretization was presented by [73]. Tests conducted on the FVC2000 DB2-A database show encouraging outcomes: 1.25% for FAR, 1.15% for FRR, and 2.83% for EER. The restricted number of photos per identification class presents challenges, such as precisely identifying reference points and producing a bigger training set. An effective cancelable biometric authentication framework using a Genetic Encryption Algorithm (GA) for increased security is presented by [74]. The process entails using GA for increased security, choosing the best sub-images, and permuting biometric templates. Promising findings were obtained via evaluation on a variety of datasets, such as the face and fingerprint datasets; nevertheless, more validation on bigger and more diverse datasets is required to ensure generalizability. A cancellable biometric security system using sophisticated chaotic maps to improve fingerprint identification was provided by [75]. Using various chaotic maps, chaos-based picture encryption produces convolution kernels that are then used to construct encrypted biometric templates. With an EER of 0.593% and a high detection probability of 96.139%), the augmented quadratic map 3 performs best in the system. The invertibility of biometric transformations and the requirement to investigate cancellable recognition in alternative biometrics are among the limitations. The Cancelable Biometrics Vault (CBV) was introduced by [76], who used winnowing and chaffing to create safe biometric templates for cryptographic key encoding. Experimental results demonstrate that, regardless of key size, the CBV’s use of an extended BioEncoding scheme is acceptable for bit strings such as iris-codes, and that its decoding accuracy is equivalent to that of the underlying CB construct. The reliance on an appropriate CB scheme for the biometric representation and computational complexity unfit for real-time applications are among the limitations. A multi-biometric cryptosystem employing Modulus Fuzzy Vault, augmenting input data with BDPHE and segmenting pictures with MBIRCH, was proposed by [77]. Features are extracted by the bidirectional deer hunting optimization algorithm, fused using score level and normalized feature fusion, and then safely stored. The approach enhances revocability and security while producing superior ROC, FAR, FRR, and GAR metrics; nonetheless, it has drawbacks related to algorithm focus and dataset authenticity. A machine vision gait-based biometric cryptosystem using a fuzzy commitment strategy was invented by [78]. Robust bits are chosen for the fuzzy commitment scheme after gait characteristics are retrieved using LTP and GEl. Tested on the CASIA A and CMU MoBo datasets, obtaining 0% FAR and FRR under certain scenarios. One limitation is that it might be challenging to identify people with complicated backgrounds or occlusions. An asymmetric cryptosystem integrating the elliptic curve method and optical scanning cryptography (OSC) was proposed by [79]. Using ECC from biometric pictures, the approach encodes things into holograms, guaranteeing excellent decryption accuracy and key sensitivity. The technique has to be optimized for speed and handles vulnerabilities to ciphertext-only assaults, even though security and key management have improved. In order to pre-process biometric images, the authors in [80] promoted the use of a fuzzy extractor that leverages deep learning, together with code-based cryptosystems to generate robust keys utilizing face biometrics. Promising accuracy metrics are found while evaluating the LFW and CelebA datasets. There is need for improvement in data privacy and security, as evidenced by limitations such as storage requirements and attack vulnerability. ElGamal encryption and Shamir’s secret sharing are two cryptographic algorithms that are used in the blockchain-based user re-enrolment approach for biometric authentication systems proposed by [81]. Key security characteristics are satisfied by simulations that show secure re-enrolment in a matter of seconds; nevertheless, scalability and adversarial assumptions are acknowledged limits. More research is needed to determine the implementation specifics and practical viability. Based on Lagrange’s interpolation, the authors in [82] provide a secure multi-biometric template protection method that uses irreversible and unlinkable image transformation. By utilizing feature-level fusion, the technique combines the properties of the fingerprint, iris, and palm print, attaining a high accuracy of 99.9816% when applied to high-resolution picture datasets. Issues include growing database sizes and possible departures from ISO/IEC 24745 standards, and privacy protection is still a problem. Biometric attributes can be randomly shuffled using a 3D chaotic map, as proposed by [83], to create a safe cancellable biometric cryptosystem. Larger datasets and other attacks are needed for more re- search, however evaluation on the ORL, FVC, and LFW datasets demonstrates encouraging results with very low Equal Error Rates (6.2460x10¹3) and high average Area un- der the ROC curve (0.9998). A robust cancellable biometric authentication technique is introduced by [84], which makes use of DNA sequencing theory, PWLCM, logistic map, and 3D chaotic maps. Through dispersion and confusion, the technique produces biometric patterns that are completely undefined while achieving increased security. Although computational complexity is still a barrier, evaluation on a variety of face and palm print datasets shows encouraging results in terms of AROC, FAR, DH, SSIM, and PSNR. An Enhanced Biometric Cryptosystem (BCS) using iris and ear modalities with Binary Robust Independent Elementary Feature (BRIEF) was introduced by [85]. Although vulnerable to database-level assaults, experimental assessment on AMI and UBIPr databases shows better performance across several parameters compared to state-of-the-art approaches, suggesting its applicability for security applications. Using deep learning and crypto-mapping, the authors in [86] provided a multi-biometric secure-storage technique that generated cancellable biometrics from fused pictures of the face, fingerprint, iris, and palm. While ICUB results reveal greater error rates, experimental results on the CASIA V4, MICHE, and MobiFace datasets demonstrate good AUCROC and low EER. The limitations mentioned by the authors include the necessity to strike a compromise between security and recognition performance, hardware requirements, and potential over fitting. In an effort to improve security, the authors in [87] proposed a multi-round AES cryptosystem with hierarchical hardware pipelined structures and biometric key generation. The system outperforms traditional AES systems in terms of space and energy efficiency because to resource sharing and simultaneous XOR operations. The undefined dataset size and the restriction to area and energy efficiency measurements are among the limitations.

Biometric characteristics are combined with cryptographic methods in biometric cryptography systems to protect authentication procedures, guarantee the privacy of biometric data, and prevent unwanted access. However, if the data is hacked, the intrinsic uniqueness of biometric features presents privacy problems. Cancellable biometrics are useful in this situation. In order to reduce the risk of prolonged exposure in the event of a data breach, cancellable biometric approaches convert biometric data into a non-reversible, pseudonymous form that can be updated or revoked as needed. The security architecture gains an extra layer of defence that improves user privacy and adaptability by incorporating cancellable biometrics into biometric cryptography systems. This addresses the reliability of authentication as well as the flexibility of maintaining bio- metric identifiers. This interaction emphasizes how crucial it is to combine cancellable biometric techniques with cryptographic security to offer a thorough and reliable solution to biometric data protection.

2.2.3. Cancellable Biometric

Cancellable biometric templates are introduced to enhance the security of biometric authentication systems by storing irreversible transformed versions of templates rather than original biometric templates. This approach prevents attacks and vulnerabilities while maintaining certain advantages, such as non-revocability. Techniques like mixing mechanisms and many-to-one transfer functions are commonly employed for generating cancellable biometrics. A basic block diagram of cancellable biometric is represented in Figure 5 where template is generated using transfer process, if by any reason the template get compromise then it is very easy to replace the old template with a new one by changing the transformation process. Retaining identifiable biometric information despite transformation efforts aimed at unlinkability, potential mismatches due to lack of adaptation to temporal changes, increased computational demands, privacy risks from potential reconstruction, and usability issues regarding recognition reliability are some of the challenges associated with using high-quality initial biometric images for generating cancellable templates. In order to guarantee the security and usability of cancellable biometric systems in actual applications, it is imperative that these variables are balanced.

A fingerprint template protection technique utilizing fused feature-level structures that produce cancelable templates was presented by [88]. Improved performance was seen with EER ranging from 1.6% to 17.6% in the FVC 2002 and 2004 datasets. Two drawbacks are the expense of computing and the invertibility brought on by fusion. In order to take advantage of flaws in partial fingerprint- based authentication systems, the authors in [89] proposed creating “MasterPrints,” which are artificial or actual partial fingerprints that match recorded templates for a large number of users. Their method proved that it was possible to impersonate users when tested on the FingerPass DB7 and FVC2002 DB1-A datasets; in some situations, Synthetic MasterPrints performed better than Sampled MasterPrints. Limitations include imbalanced datasets and limited application to minutiae-based systems, which have prompted more study into remedies. A one-factor cancellable bio- metric authentication technique using Indexing First Order hashing was suggested by [90]. It was assessed for ac- curacy performance, non-invertibility, renewability, and unlinkability. Tested on six datasets from the FVC 2002 and 2004 databases, yielding metrics for genuine-imposter distribution and Equal Error Rate (EER) that are adequate. Although it separates IDs from biometric templates, it assures unlinkability and is susceptible to standard symmetric key cryptosystem assaults. By using OIOM hash and MSH to create a pseudonymous identification, the authors in [91] promoted a one-factor cancellable palmprint recognition technique that achieved a recognition accuracy of 98.07%. although no specific limitations or restrictions are mentioned, experiments were conducted using the PolyU and TJU palmprint databases. A global multi-biometric system using deep neural networks for cancellable feature creation was proposed by [92], which achieved good performance on iris datasets (IITD Iris and MMU2). The technique acknowledges constraints in sensitivity and adaptability and blends dimensionality reduction, adaptive fusion, and revocability. Aspects of security and privacy are explored, and some adversarial threats are illustrated. In order to provide maximum safety of sensitive data, the authors in [93] provided the Secure Triplet Loss method for training end-to-end deep learning models to build non-invertible and unlinkable biometric templates. The approach is tested on facial recognition and ECG tasks, showing that it can successfully modify pre-trained models or create secure models from scratch. The dataset utilized for experimentation is not stated, despite constraints being addressed, such as sensitivity to certain assaults and the effect of demographic characteristics on system accuracy. Constrained Optimized Similarity-based Attack (CSA) was introduced by [94] as an improvement over earlier similarity-based assaults. It incorporates algorithm-specific limitations to optimize pre- image production for impersonation. The usefulness of CSA is demonstrated through experiments using the labelled Faces in the Wild (LFW) dataset and Index-of-Max (IoM) hashing. The performance measures that are measured include the total success rate, the False Acceptance Impostor (FAI) rate, and the True Acceptance Impostor (TAI) rate. The success of CSA is contingent upon data leakage and constraint identifiability, which may restrict its application in certain biometric systems. In order to prevent unwanted access, the authors in [95] developed a non-invertible cancellable fingerprint template approach based on Delaunay triangulation of minutiae points. When evaluated against the FVC2002 database, the findings show promise in terms of identification accuracy, resilience against fingerprint distortion, and comparison to modern approaches. However, more research is necessary for wider validation due to concerns including hacked acquisition equipment and dataset size restrictions. A safe method for building fingerprint templates that are optimized was disclosed by [96]. This technique uses the quality of minutiae points to generate 3D shell-shaped templates. Tested on nine fingerprint datasets, the method yields a 0% Equal Error Rate (EER) and 100% accuracy in separating authentic individuals from imposters. Nevertheless, low-quality photos could be limiting its performance, indicating a possibility for improvement through multimodal biometric system integration. Using IoM hashing and BioHashing, the authors in [97] developed the first constrained-optimized similarity-based attack (CSA) against cancellable biometrics (CB). By optimizing preimage creation through algorithm-specific limitations, CSA outperforms Genetic Algorithm enabled similarity-based assaults (GASA). Success rate (SAR) and false acceptance index (FAI) of CSA, assessed on the LFW dataset, demonstrate the efficacy of CSA in breaking IoM hashing and BioHashing security. However, there are drawbacks to CSA due to its constant model complexity and dependence on hash code size. A cloud and Internet-of-things-ready method for cancellable biometric template generation was developed by [98]. By using the Greatest Common Divisor (GCD) between hazy biometric pictures, it guarantees authentication accuracy and non-recoverability. It is tested on several biometrics and achieves low EER values (down to 0.04%) and high AROC values (up to 99.59%), while it is recognized to have limits with respect to pre-defined distortions and picture quality. A cancellable biometric authentication system with feature-adaptive random projection that includes feature extraction, transformation, and encrypted domain matching was presented by [99]. EER, GAR, and FAR metrics evaluation on the FVC2002 and FVC2004 databases revealed competitive performance. The need for more dis- criminating feature descriptors and investigation across a range of biometric features are acknowledged constraints. In order to provide biometric data safety and reliable authentication, the authors in [100] suggested a user authentication and key agreement system utilizing cancellable biometrics and PUF in multi-server settings. Their methodology beat state-of-the-art techniques when tested on the LFW dataset and assessed using the CMC, ROC, DIR, and DET curves. Recognized drawbacks include the requirement for assessment against complex assaults and scalability issues with larger data sets, indicating potential directions for further study. BioCanCrypto, a biocryptosystem using fingerprint cancellable templates that combines biocryptosystems and cancellable biometrics, was proposed by [101]. It takes cancellable templates and extracts cryptographic keys using a reusable fuzzy extractor with LDPC coding. tested using the FVC2002 dataset, showing encouraging outcomes despite low FRR and EER. The study is restricted to fingerprint data, though, thus it might be worthwhile to investigate alternative feature spaces or data modalities. A unique biometric template protection approach incorporating watermarking and cancellable transformation was suggested by [102]. The technique uses pair-polar coordinates with cancellable modification of minutiae and binary watermarking obtained from fingerprint minutiae. tested on the BioSecure and FVC2002 DB1 datasets, showing better EER and resilience to assaults than previous approaches. Investigating several biometric modalities to improve accuracy and robustness is one of the future research priorities. Absolute Value Equations Transform (AVET), a non-linear projection technique that ensures irreversibility by depending on the Absolute Value Equations issue, was promoted by [103]. AVET surpassed state-of-the-art approaches in bimodal circumstances and produced competitive performances in unimodal settings after being evaluated on eight datasets for different biometrics. A fixed sample size and susceptibility to brute-force attacks are two limitations. Using deep fusion and deep dream, the authors in [104] propose a multi-biometric cancellable system (MBCS) that creates tamper-proof templates using fingerprint, finger vein, and iris biometrics. It used a dataset with nine pictures per modality, outperforming comparable algorithms in EER, FAR, FRR, and AROC. All quantitative evaluations showed positive results. Larger datasets, high computing demands, and a drawn-out enrolment procedure are among the acknowledged limits. A cancellable multi-biometric identification system using ACM encryption and decimation to combine biometric data into a single template was presented by [105]. Performance measures, which were assessed using FVC 2002 and ICE 2005, included EER and ROC curve analysis. Its stated drawbacks were the need for better parameter estimate techniques and bigger databases. CSMoFN is a revolutionary cancellable multimodal biometrics system that combines face and periocular data, as reported by [106]. It uses pairwise angular loss and ArcFace for training, and it achieves an EER of 6.67% on average across six datasets and 2.12% on Facescrub. The authors warn against potential CB template inversion concerns while highlighting dual template-changeability and acknowledging difficulty in fair comparison. A new cancellable multi-biometric system combining deep learning- based fusion and selective encryption was developed by [107]. AES encryption, PRNG matrix XORing of a chosen ocular picture, and Viola-Jones facial segmentation are all used. Although constrained by the size of the dataset, it surpasses previous efforts with high entropy and low correlation, quicker enrolment, and better metrics like EER and AROC, indicating that medical imaging application is a promising area for future study. A unique cancelable biometric solution using deep learning for fingerprint and face biometrics on smart phones was presented by [108]. It improves security for Internet of Things applications by using Siamese networks with visual and text encoders. Prolonged testing shows improved performance compared to earlier approaches, surpassing ciphering-based systems in EER, FAR, FRR, and AROC, and attaining high accuracy metrics. MBFH, a unique cancellable biometric method that makes use of safe hashing and non-invertible transformations, was invented by [109]. Multi-biometric acquisition, feature ex- traction, Multi-Exposure Fusion (MEF), and SHA-3 hashing are all included in the process. Tests using retina, finger veins, palm, and dorsal vein pictures show high pairwise distances and good performance in generating text and visual templates, indicating flexibility for further improvements such as adding white Gaussian noise. A biometric template protection system for Euclidean and Cosine metrics was introduced by [110]. It used distance-preserving, one-way, and obfuscation modules with location-sensitive hash functions. Effectiveness against similarity-based and linear inequality attacks is demonstrated by evaluation on face datasets such as AR, CASIA, ORL, and LFW. However, this approach is not always the best option and requires pre-processing, which raises the computational costs significantly. Despite defences against assaults, the authors advise against assuming complete security in any situation. Using the chaotic Baker map to encrypt biometric templates, the authors in [111] provide a chaotic-based cancellable face recognition system that is highly accurate (98.43%) and adaptable to a variety of databases. The approach takes care of lighting, occlusion, and emotions, but it also recognizes the necessity for environmental adoption and trade-off between performance and user privacy. A cancellable biometric authentication system using image style transfer is proposed by [112]. During registration, users supply a face picture and a key image to create and store a template. In order to get high AUC values (>0.9) in most circumstances, authentication entails comparing freshly created templates with stored ones. Correlation coefficients and ROC curves are used to evaluate the results. One of the limitations is the possibility of authentication using marginally similar key images. This suggests that future research should put limitations on key image parameters.

Widely utilized in forensic and law enforcement settings, AFIS faces unique security issues such data breaches, spoofing attempts, and system flaws that could jeopardize the integrity and accuracy of fingerprint matching. In a similar vein, these vulnerabilities also affect general-purpose user authentication systems, which use fingerprints for identity verification and access control in common applications. Consumer-grade fingerprint authentication systems can benefit from and benefit from the protective tactics included in AFIS security measures. Developers can better address similar risks in general-purpose systems by knowing the vulnerabilities and countermeasures relevant to AFIS. This will help to ensure that strong security mechanisms are put in place to protect user data and prevent unwanted access. In order to improve overall security and user trust in biometric technology, it is crucial to transfer high-security solutions from specialized applications to more commonly used fingerprint authentication systems.

Using global models, local analysis, and combination approaches, the authors in [113] provide techniques to improve fingerprint orientation extraction, obtaining an EER of 0.206% on the FVC2006 DB2 dataset. In addition to photos and has difficulties identifying pores in occluded or blurred images, the system outperforms previous algorithms and achieves excellent accuracy across touch-based, touch less, and latent fingerprint datasets. Future research may concentrate on creating matching algorithms for diverse fingerprint photos using derived pore properties. A new technique for high-resolution fingerprint pore comparison using rotational invariant edge descriptors and Root-SIFT descriptors was introduced by [2]. EERs and FMR1000s for the FVC2002 datasets are 1.86% and 0.12%, respectively, according to the evaluation. The authors support direct pore comparison methods in spite of the possible loss of actual correspondences, acknowledging the drawbacks of comparing pores only on the basis of geometric distance. In order to improve ridge structures and remove false minutiae, the authors in [121] provide a fingerprint enhancement and reconstruction method based on orientation and phase reconstruction. Despite limitations such as age, dirt, and medical issues impacting picture quality, validation on the FVC2002 and FVC2004 datasets shows promise for enhancing interoperability among minutia encoders and matchers. DeepPoreID was created by [122], which achieved higher performance, especially with tiny picture sizes, by using deep convolutional networks for descriptive pore characteristics in fingerprint matching. Significant gains in EER and FMR100 are shown by evaluation on high-resolution fingerprint databases; nonetheless, there are certain drawbacks, such as a need for a sufficient number of pores and difficulties with occlusion. It is underlined that high-resolution finger- print databases are available as open-source resources for more thorough testing. A homomorphic encryption-based fingerprint authentication system that guarantees access control while protecting biometric template data was investigated by [123]. Tested on FVC2002 DB2, the system shows an EER of 9.23%. The computational time issues that were found encourage research into more effective homomorphic encryption techniques. For a thorough assessment, more testing with bigger datasets and varied scenarios is advised. The Shark Smell Optimization (SSO) technique was used by [124] to study a fingerprint authentication system. Achieved FAR of 0.00%, FRR of 0.00666%, and CVR of 99.334% after evaluation on a dataset of 150 student fingerprint pictures. The size of the dataset, lack of variety, and exclusive dependence on the SSO algorithm are limitations. Future research into more swarm algorithms and bigger datasets is advised to get better results. Using fully homomorphic encryption (FHE) and the TFHE library, the authors in [125] created an effective fingerprint authentication system that preserves privacy while guaranteeing safe processing and storage of fingerprint data. Tested on NIST Special Database 9, the system reaches fingerprint matching in an average of 166 seconds, acknowledging the time consumption of the bootstrapping procedure and suggesting improvements for the future. BioSec, a biometric authentication system using fingerprint authentication for secure communication among edge devices, was introduced by [126]. AES-128 bit encryption is used to protect biometric templates while they are being sent and stored in databases. Assessed using the FVC-2004 DB3 dataset with 70% accuracy, they recognize the limitations of symmetric encryption vulnerability and propose improvements to privacy mechanisms, study of lightweight encryption methods, and investigation of other biometric markers such as iris for enhanced security and usefulness. In [127], the authors invented a blind and reversible fingerprint image watermarking method using the differential method and DCT domains. Assessed on the FVC2002 fingerprint database, the technique employs DCT-transformed sub-vectors to incorporate watermark bits, facilitating direct watermark access throughout the extraction procedure. Evaluation measures included matching score and PSNR, which demonstrated a respectable level of fingerprint image security maintenance. The authors pointed out drawbacks like as noise sensitivity and compression, indicating that a bigger sample size would be necessary to increase performance. A set of two finger- print matching methods, employing mtriplets and cylinder codes, fused with a supervised classifier for latent finger- print recognition, was introduced by [128]. Achieved Rank-1 identification rates of 74.03% and 71.32% were evaluated on NIST SD27, GCDB, and MOLF DB4 databases. This work improves upon earlier approaches by evaluating matches using cumulative match characteristic (CMC) curves and plans to investigate other variables in further research for more accurate matching characterisation. Using attention mechanisms and Monte Carlo drop-out, the authors in [129] improved upon previous approaches by introducing an explainable fingerprint ROI segmentation model. High Jaccard similarity and Dice score measurements were obtained by testing on FVC datasets. Limitations include misclassification of back- ground noise and performance changes caused by the size of the sample. An unsupervised technique for evaluating fingerprint quality based on minutiae detection confidence is called MiDeCon, and it was proposed by [130]. tested on FVC 2006 datasets; verification error and error-vs-reject curves showed superior performance compared to NFIQ1 and NFIQ2. The study did not address the limitations and limits of MiDeCon. ASRA, an Automatic Singular Value Decomposition-based Robust Fingerprint Image Alignment technique, was proposed by [131]. When tested on the FVC2002 and FVC2004 databases, ASRA demonstrated efficacy and efficiency in alignment, but with recognized shortcomings in terms of picture backgrounds and ROI extraction. For approach efficacy, more validation on bigger datasets is recommended. A unique fingerprint template protection and authentication method utilizing visual secret sharing (VSS) and super-resolution was suggested by [132]. The approach demonstrated security and resilience when tested on the FVC2002 DB1 dataset, while there are some acknowledged drawbacks, such as the requirement for higher-quality photos and the possibility of attacks on the super-resolution process. A dual-filter architecture and a novel texture descriptor, sDSIFT, were recommended by [133] to improve spoofing detection in fingerprint authentication systems. Tested using LivDet 2013 and 2015 datasets, the approach combines the dual-filter architecture with sDSIFT to achieve competitive accuracy. Nevertheless, shortcomings include the incapacity to distinguish between different filter kinds and spoofs created using materials that resemble real fingerprints. Three reliable methods utilizing fingerprint, iris, and voice characteristics for multi-modal biometric authentication were presented by [134]. The second technique, which used an SVM classifier and sum FFC, was evaluated on a dataset of 228 image and signal data, and it obtained a 100% classification rate; however, its scalability and real-world resilience were not evaluated. A secure online fingerprint authentication system with a cancellable fingerprint template design for privacy was provided by [135] for Industrial IoT devices over 5G networks. Tested on six public fingerprint databases, the system demonstrates competitive efficiency and performance, however it recognizes the difficulties in preserving security while guaranteeing strong authentication over picture databases of poor quality. Using an Arduino Mega 2560 and a fingerprint sensor, the authors in [136] developed a fingerprint authentication system based on embedded systems. It generates templates using minutiae extraction and provides PIN and fingerprint combinations for smart home access. Although the accuracy is said to be high, low-end embedded systems’ cost and durability against assaults present obstacles. CNN classification and Gabor filter-based feature extraction were used by [137] to investigate a fingerprint authentication technique. The approach beats other modern methods on the FVC2006 database, with a validation accuracy of 99.33%. Smaller training sets are still difficult to manage, despite results. A deep CNN model for fingerprint authentication was studied by [138], with 100% training and validation accuracy. One limitation is the diversity of datasets, which raises the possibility of strengthening robustness and generalizability. A multi-factor authentication approach was created by [139] to reduce False Acceptance Rates (FAR) and False Rejection Rates (FRR) in biometric systems. By combining several fingerprint sample permutations, their method increases system accuracy and fortifies it against impostor assaults. As opposed to single-factor authentication, experimental results on datasets like CASIA-FingerprintV5 and FVC2002 show lower mistake rates; nonetheless, the study admits several limitations, including addressing environmental deterioration and dataset size. InfinityGauntlet, a brute-force attack against smartphone fingerprint authentication (SFA) systems, was revealed by [140]. It uses SPI MITM to circumvent limit attempts and takes use of design flaws. They tested the assault on several devices using inexpensive equipment and a synthetic fingerprint generating technique, and they were able to detect four different types of attacks. One limitation is that security regulations demand the attack to be finished within 72 hours and that sensor hot plugging support be provided. A unified model for fingerprint authentication and spoof detection was developed by [141] using DualHeadMobileNet, a dual-head convolutional neural network. The model, using integrated datasets from FVC 2006 DB2A and LiveDet 2015, delivers low spoof detection error rates and good authentication accuracy. Computational effort and the requirement for bigger, more varied datasets for further studies are among the limitations. Blind-Touch, a privacy-preserving fingerprint authentication system using homomorphic encryption and machine learning, was introduced by [142]. Using PolyU and SOKOTO as benchmark datasets, the system attains great accuracy, with F1-scores of 93.6% and 98.2%, respectively. The influence of feature vector size on calculation time and the computational cost of constructing a traditional CNN with homomorphic encryption are among the challenges. The FingerPIN method, which combines PINs with fingerprints for multi-factor authentication, was proposed by [143]. According to a usability research, FingerPIN is more secure, quick, user-friendly, and efficient than conventional techniques. Restrictions include a limited sample size and a brief period of data collection, indicating the need for more study. An AVAO-enabled Deep Maxout Network (DMN) for fingerprint-based person authentication was proposed by [144]. The technique obtained an accuracy of 0.927 on the CASIA Fingerprint Image Database by employing the AVAO optimization algorithm of the method. The authors pointed out constraints such as the need for strong preprocessing and model training.

Fingerprints are being utilized more and more in general- purpose user identification to confirm identity on a variety of platforms, including secure facilities and mobile devices. This calls for quick and accurate verification procedures. Rapid fingerprint indexing and searching technologies are essential for fulfilling these expectations because they allow fingerprint records to be retrieved and matched quickly, improving system performance and user experience. The time needed for fingerprint verification can be greatly decreased by integrating sophisticated indexing and searching algorithms into general-purpose authentication systems, increasing the process’s efficiency and scalability. This synergy emphasizes how crucial it is to optimize fingerprint search technologies to facilitate secure and seamless user authentication, guaranteeing that the speed and accuracy of the underlying verification procedures meet the convenience of biometric access.

4. Fast Fingerprint Indexing and Searching

In various application domains, the rapid response and high efficiency of an Automated Fingerprint Identification System (AFIS) are paramount. To achieve swift results from large datasets, a common approach involves using a combination of fast and accurate algorithms. This entails employing a fast algorithm to generate a subset of candidate fingerprints, followed by a more precise but slower algorithm for final matching. Efficient fingerprint indexing predominantly revolves around two strategies: (i) utilizing Level-I features like ridge orientation and frequency maps, and (ii) leveraging minutiae features. The primary aim of fingerprint indexing is to reduce search space and time, especially when dealing with massive gallery image datasets. This section highlights several algorithms that exemplify this approach. Figure 7 shows the fingerprint processing and matching in distributed environment where the features from the fingerprint image is extracted at the client side and passed it to a server, where multiple feature matchers run in parallel to match the query feature in large fingerprint database. After matching the result is again send back to the client side.

To analyse huge databases efficiently, the authors in [145] created a distributed fingerprint matching system. The system achieves flexibility and scalability by using a two-level distributed design. Studies using NIST and synthetic fingerprint datasets showed that, when processing nodes and thread counts rose, execution times improved while accuracy remained similar to that of conventional AFIS. By parallelizing database searches, the authors in [146] introduced a GPU-based fingerprint recognition system that achieves fast processing rates. Although the approach can quickly identify millions of fingerprints per second, its practical use is limited by its 2% mistake rate. While highlighting the promise of GPU-based algorithms, the paper also notes that more development and integration into hybrid systems are required to achieve higher accuracy.

Using locality sensitive hashing (LSH) and minute cylinder-code SDK for effective indexing with pose limitations, the authors in [147] developed an absolute registration method for fingerprint indexing. Experiments using databases such as the NIST Special Database 14 and a law enforcement fingerprint database shown increased efficiency and accuracy. Among the drawbacks are the high dimensionality of minutiae descriptors and the requirement for additional study to improve resilience against noise and un- predictability. In [148], the authors presented a fingerprint indexing technique that uses convex core points and minutiae pairs, combining candidate list reduction and k-means clustering to achieve effective indexing. The methodology outperforms state-of-the-art methods on six datasets by using coaxial Gaussian track coding and MBP representation for precise matching. One limitation is that the hardware and software utilized in the trials varied, and there is a lack of information on the values of the parameters.

Improved retrieval efficiency was achieved by [149] using deep convolutional neural networks (DCNN) to build a minutiaecentred fingerprint indexing approach. They achieved superior results on five benchmark databases with their unique aggregating methodology that made use of 1-D CNN. Metrics for evaluation included penetration and error rates; higher penetration rates and lower error rates were noted. Two of the limitations were the high processing costs for MCC-based approaches and the low quality photos in FVC2000 DB3a. To address the shortcomings of minutiae-based approaches, the authors in [150] studied DeepPrint, a deep network for obtaining fixed-length fingerprint representations. Using the NIST SD4 and SD14 datasets, DeepPrint outperformed leading commercial off- the-shelf (COTS) SDKs in terms of accuracy and search speed. DeepPrint showed encouraging results in finger- print identification, despite limits in benchmark saturation. CNNAI, a CNN-based technique for fingerprint recognition utilizing geometrical minutiae arrangements, was studied by [151], and it outperformed state-of-the-art algorithms in terms of identification rates. CNNAI demonstrated its abilities by achieving Rank-1 identification rates of 84.5% and 80% on the FVC2004 and NIST SD27 latent fingerprint datasets, respectively. Reliance on a minimum of eight minutiae points is one of the limitations, and using MINU- EXTRACTNET to extract all minutiae presents difficulties that might potentially impact identification rates. By developing a pore-based indexing technique for high-resolution fingerprints, the authors in [152] significantly reduced pre- selection error rates compared to previous approaches on a variety of databases. The approach outperformed state-of- the-art minutiae-based algorithms on DBI and IITI-HRFP, according to tests conducted on datasets comprising DBI, DBII, IITI-HRFP, and IITI-HRF. Restrictions include decreased efficacy on databases with high-resolution whole fingerprints, which motivates more study into distortion correction methods. A latent fingerprint indexing technique integrating minutiae-based features, global and local matching, and machine learning-based segmentation was suggested by [153]. Up to 94.77% accuracy was shown in experiments on the IIIT- D dataset, with indexing times varying from 5.29 to 25.92 seconds. Managing incomplete perceptions, background noise, inadequate ridge clarity, and nonlinear distortions are among the difficulties. An MCC-based indexing technique for latent fingerprint recognition was described by [154], beating competitors on the NIST SD4 and NIST SD14 databases by at least 1% and 3%, respectively. Although effective, it can have performance issues with different kinds of databases or with less-than-ideal detail extraction, and its temporal complexity prevents real-time applications. Employing graph-based algorithms for indexing and refining, the authors in [155] provided a pore-based fingerprint retrieval technique that outperformed previous approaches on databases DBI and DBII in terms of speed and accuracy. Its shortcomings, however, include the dependence of indexing precision on cluster accuracy and the computational cost that increases with the number of pores. Two fingerprint search strategies utilizing classic inverted index methods were proposed by [156]. They were examined on FVC2002 DB1a and a private dataset, with differences in document generation methodologies and minutiae handling. The second approach’s scalability and sensitivity to low-quality minutiae were found to be limits, despite the fact that it produced reduced error rates. Using highly discriminative embeddings for constant-time identification and k-means or LSH index table generation, the authors in [157] present PalmHashNet, a palm print database indexing technique. Evaluation on four benchmark databases shows above 99% accuracy; nonetheless, there are several constraints, such as assumptions about the positioning of the palm area and unproven scalability on big datasets. By combining GIST descriptors from face, iris, and palm print biometrics, the authors in [158] developed an efficient multidimensional spectral hashing (MDSH) technique for data retrieval from multi biometric databases. Despite computational and memory constraints, experiments conducted on databases from IIT Delhi reveal enhanced accuracy measures that surpass tree-based indexing approaches and random hashing techniques. More efficiency improvements are proposed, such as lower memory use and processing costs, and investigating alternatives to local features. To lessen the computational burden in biometric identification systems, the authors in [159] proposed a nearest quality score-based intelligent search, which resulted in workload reductions of up to 38% for face, 31% for iris, and 29% for fingerprint databases. The technique, tested on the FERET, CASIA, and FVC2002 datasets, effectively indexes queries using sample quality scores that may be applied to a wide range of biometric features. Intraclass variability and susceptibility to various quality estimator techniques affecting the number of comparisons are challenges. FKPIndexNet is an innovative approach for finger-knuckle-print (FKP) identification, developed by [160]. It uses customized autoencoder networks and similarity-preserving hash codes. The PolyU-FKP and IITD FKP databases show 100% hit rate in experimental findings at low penetration rates, indicating that index table creation approaches have limitations. A signal-based fusion approach for indexing biometric databases was presented by [161], which reduced computational effort by up to 70% without sacrificing biometric performance. Promising results are obtained when evaluated on benchmark databases such as FERET and FRGCv2, however these need the recomputation of indexes for new enrolments and high-quality face photos. By utilizing cancellable methods and DNN- based embedding extractors, the authors in [162] developed a novel approach to multi-biometric indexing that preserves privacy while improving biometric performance by 57%. Evaluation using a composite dataset reveals advancements but also points up extractor compatibility and dataset representation constraints. In their investigation of AFR-Net, an attention-driven fingerprint recognition network, the authors in [163] combined CNN-based and attention-based embeddings for improved performance. By utilizing a variety of datasets, AFR-Net surpasses baseline models and attains excellent accuracy metrics in verification and identification tasks. The lack of domain expertise in fingerprints and the requirement for better latent fingerprint preparation methods are among the limitations. An effective Gravitational Search Decision Forest (GSDF) method for fingerprint identification was created by [164] by merging the gravitational search algorithm (GSA) and random forest (RF). Comparing the GSDF technique to conventional machine learning classifiers, it yields higher identification rates for both entire and latent fingerprints. The authors agree that although the technique beats current methods, additional tuning is necessary for the recognition of low-quality latent fingerprints.

Infant fingerprint identification presents a unique set of challenges compared to adult fingerprint identification, which has a well-established sector with reliable techniques and tools for collecting and evaluating mature fingerprint patterns. Because of their fingers’ quick growth and development, infants’ fingerprints are more prone to distortion and have less defined ridges, therefore effective identification requires specific methods and modifications. It is necessary to modify fingerprint capture and processing technologies that have been developed for adults in order to address these issues with newborns. Researchers and practitioners can improve the overall efficacy and accuracy of fingerprint identification systems across all age groups by linking the methodology and technology utilized in both fields. This will guarantee that the systems remain accurate and dependable from infancy through adulthood. For comprehensive fingerprint identification systems that meet the needs of a wide range of demographics, this integration is essential.

5. Infants’ Fingerprint Identification

Emerging as a challenging application, infants’ fingerprint identification plays a crucial role in maintaining accurate records of a child’s vaccination history, nutritional needs, and overall identity throughout their lifetime. Among various biometric options, fingerprinting emerges as one of the most suitable choices due to its stability and uniqueness. While alternative biometrics like iris scanning or facial recognition are less feasible for infants due to practical challenges, fingerprints offer a reliable biomarker. In the field of infants’ fingerprint identification, limited research exists, but several notable contributions have been gathered from the literature. Figure 8 shows fingerprints of an infant at different age, which demonstrate the different density of ridges during different age of a infant.

A cost-effective, high-resolution fingerprint scanner for newborns was created by [165]. They achieved a rank-1 accuracy of 73.98% by building a database with pictures from 16,384 infants. In [166], the authors assessed the accuracy of child fingerprint recognition using NFIQ 2.0 measures in a follow-up research. Using a 500 PPI dataset, they were able to achieve TAR of 99.5% with FAR of 0.1% for children aged 12 months. An inexpensive newborn fingerprint identification system called InfantPrints was created by [167]. It consists of a high-resolution matcher and a specially designed fingerprint scanner. Assessed on a 315 baby longitudinal database, the system demonstrated excellent accuracy from enrolment at 2 months to authentication at 1 year. Among the drawbacks is the lack of a reliable automatic alignment technique for baby fingerprints. Using wavelet feature extraction and K-NN classification, the authors in [168] developed a baby’s footprint identification system. Using a 200×500 pixel ROI and level 4 wavelet decomposition, the system obtained 99.30% accuracy with a dataset consisting of 600 footprint photos and 30 newborns. The modest sample size and the requirement for more testing on a bigger dataset for real-world application are acknowledged constraints. A newborn fingerprint identification technique using deep learning for fingerprint categorization was described by [169]. The approach obtained 78.4% classification accuracy compared to hand classification with a dataset consisting of 1,357 training photos, 166 validation images, and 1,181 test images. The requirement for significant computer resources and the difficulties in obtaining clear fingerprint pictures from some newborns are among the limitations. Using fingerprint, iris, and outer ear shape modalities, the authors in [170] created and assessed biometric recognition systems for babies. For the fingerprint and iris modalities, new hardware and soft- ware were created, but the outer ear shape modality made use of already-existing technologies. At a public clinic, data on newborns and young children under the age of one was gathered. FTA (failure to acquire) and equal error rate (EER) were used as accuracy measures. With an EER of 0.16%, the ear modality performed the best; nevertheless, inadequate reporting, inconsistent devices, and data collection from newborns and young toddlers presented difficulties. A multi-instance contingent fusion approach for newborn fingerprint verification was presented by [171], which combined the left thumb and right index finger prints of infants ranging in age from one day to six months. The approach achieved 73.8%, 69.05%, and 57.14% accuracy for 1-month, 3-month, and 6-month intervals between enrolment and query prints, respectively, by using a 500 ppi fingerprint scanner and NIST feature extractor MINDTCT. Among the limitations were the requirement for fusion with increasing time intervals between photos and the lack of supplemental information in permission forms. M²BRTPC, a new Multi-modal Biometric Recognition for Toddlers and Preschool Children, is presented by [172]. Its goal is to identify children using minimum biometric characteristics such as fingerprints, faces, and iris. Their research, which acknowledged age range restrictions and the inherent difficulties in biometric identification, improved performance by 14.75e¹8.75% using iris and fingerprint data from over 100 children ages 18 months to 4 years. Using a pre-trained ResNet-50 model, the authors in [173] investigated a fingerprint identification method for infants and toddlers, solving issues with image clarity and quality. The process entails cropping and enhancing images in order to precisely extract fingerprint traits. Using a dataset of 154 participants, the experimentation yielded an accuracy of 82.47%, a false rejection rate of less than 18%, and an authentication duration of about 2 seconds per fingerprint, while taking into account restrictions such as tiny fingerprint size and picture quality limits. Employing the CLCF dataset supplemented with a hybrid technique, the authors in [174] presented Child- CLEF, a CNN-based children detection system employing contactless fingerprints. The suggested Child-CLEF Net model outperforms current systems with 98.46% accuracy and 1.99% equal error rate by extracting minutiae and using BOZOROTH3 for identification. Larger datasets are required, as is the exploration of other enhancement and feature extraction techniques. Future developments should take patch-based newborn identification into consideration. These are some of the limitations.

Biometric identification systems for newborns and toddlers have advanced significantly in terms of accuracy and technological adaptability, but they still face significant obstacles in terms of data quality, moral issues, and practical implementation. Improving algorithms, expanding datasets, strengthening privacy protocols, and improving usability will be crucial in tackling these issues and boosting the dependability and relevance of these systems in paediatric healthcare and other fields.

A key component of biometric authentication is fingerprint identification, which uses distinctive ridge and valley patterns to confirm each person’s identity. Though useful, it can have drawbacks such spoofing, ambient conditions, and differences in fingerprint quality. In order to improve accuracy and reliability, multi-modal identification systems integrate various biometric modalities, such as voice prints, iris patterns, fingerprints, and facial recognition. Multi- modal systems are able to complement fingerprint data with other biometric identifiers, making up for any one modality’s shortcomings and offering a more robust and all- encompassing authentication procedure. This integration makes biometric systems more flexible and user-friendly while simultaneously enhancing overall security. Knowing how techniques might improve fingerprint recognition shows the possibility of developing more flexible and safe biometric systems that take advantage of several biometric properties.

6. Multi-modal Fingerprint Biometric

In [175], the author reported a multi-modal biometric system that combines face, fingerprint, and signature modalities with feature extraction techniques including Principal Component Analysis and Stationary Wavelet Trans- form. In order to address problems like spoof attacks and noisy data, they evaluate system accuracy on the YALE- FVC2002KVKR database by applying score and decision level fusion. Through the use of multi-modal strategies to overcome restrictions, the project aims to boost the dependability of biometric systems. In [176], the authors proposed improved unimodal and multi-modal biometric recognition systems using fingerprints and ECG signals using both traditional approaches and deep learning. When tested on virtual datasets like the FVC2004 fingerprint database and the MIT-BIH ECG database, multi-modal systems perform better than unimodal ones. Some of the drawbacks are the scarcity of ECG databases, the challenge of finding real- world datasets, and the high processing costs related to deep learning models. In [177], the authors have presented a deep learning-based multi-modal biometric fusion model that blends score, feature, and pixel layers to increase identification accuracy. Evaluation on a simulated dataset combining iris, fingerprint, and face data shows 99.6% accuracy using Euclidean distance metric learning and modality-specific network training for practicality. Limitations originating from the dataset’s dependability may affect the accuracy rates. A 98.5% accurate HGSSA-bi LSTM model that combines fingerprint and iris biometrics is presented by [178]. When tested on the CASIA dataset, the model demonstrates excellent sensitivity and precision, but it also acknowledges the cost and complexity of multi-modal systems. In [179], the authors offers “Secure Sense,” a multi-modal biometric system that incorporates face, fingerprint, and iris data and achieves 93% accuracy. The limitations of unimodal systems are addressed by decision-level fusion approach, which enhances strong authentication by utilizing real-time and web-based datasets.

7. Comparative Analysis

The Table 1 provides a comprehensive overview of un- formatted fingerprint image authentication methods, encompassing various techniques aimed at enhancing the quality and recognition accuracy of latent fingerprints. Each method offers unique advantages and faces specific limitations, highlighting the need for ongoing research to address existing challenges and explore new avenues for improvement. Future research in this field should focus on several key directions. Firstly, there is a pressing need to develop robust and scalable algorithms capable of handling diverse and challenging latent fingerprints, including those with low quality, partial information, or distortion. This may involve further exploration of deep learning architectures, such as convolutional neural networks (CNNs) and generative adversarial networks (GANs), to effectively enhance latent fingerprint images while preserving crucial details and minimizing noise. Secondly, efforts should be directed towards standardizing evaluation protocols and benchmark datasets to facilitate fair comparisons between different methods and promote reproducibility across studies. This includes the creation of large-scale, publicly available databases encompassing a wide range of latent fingerprint images captured under various conditions, which can serve as a common test bed for evaluating the performance of different algorithms. Additionally, research should address ethical considerations surrounding the deployment of unformatted fingerprint image authentication systems, including privacy concerns, algorithmic bias, and potential misuse of biometric data. Developing transparent and accountable frameworks for data collection, storage, and usage is essential to ensure the responsible and ethical implementation of these technologies in real-world applications. Furthermore, the integration of multi-modal biometric fusion techniques, such as combining fingerprint with other biometric modalities or contextual information, holds promise for enhancing the robustness and reliability of authentication systems, especially in challenging scenarios or under adversarial conditions. Lastly, advancements in hardware technologies, such as high-resolution sensors and efficient processing units, can significantly contribute to improving the accuracy and efficiency of unformatted fingerprint image authentication methods. Collaborative efforts between academia, industry, and regulatory bodies are essential to drive innovation, address emerging challenges, and ensure the continued advancement and responsible deployment of fingerprint authentication technologies in various domains. By pursuing these research directives, the field of unformatted fingerprint image authentication can continue to evolve and meet the growing demand for secure and reliable biometric authentication solutions.

Table 1: Comparative Analysis of Unformatted Fingerprint Image Authentication Methods

The Table 2 provides a detailed comparative analysis of various fingerprint liveness detection methods, encompassing a range of techniques such as feature extraction, machine learning, deep learning, and fusion approaches. Each method demonstrates different strengths and limitations in terms of accuracy, dataset applicability, and susceptibility to spoofing attacks. For instance, deep learning-based approaches, such as convolutional neural networks (CNNs) and stacked auto encoders, have shown promising results in achieving high detection accuracy. However, they often require large and diverse datasets for training to generalize well across different sensors and spoofing materials. Future research in fingerprint liveness detection should address several key challenges and explore new avenues for improvement. Firstly, there is a need to enhance the robust- ness and generalization capabilities of existing methods by leveraging larger and more diverse datasets that encompass various sensor types, image qualities, and spoofing materials. Additionally, researchers should focus on developing techniques capable of detecting low-resolution fingerprints and effectively handling imbalanced datasets to mitigate bias and improve overall performance. Secondly, the development of adaptive and resilient liveness detection systems that can dynamically adjust to evolving spoofing attacks is crucial. This may involve exploring novel approaches such as adversarial training, ensemble learning, and anomaly detection to detect and adapt to emerging threats effectively. Moreover, incorporating multi-modal biometric fusion techniques, such as combining fingerprint with iris or face modalities, could further enhance the robustness and reliability of liveness detection systems. Furthermore, research efforts should concentrate on addressing ethical considerations related to the deployment of fingerprint liveness detection systems, including privacy concerns, algorithmic fairness, and transparency in decision-making processes. Developing standards and guidelines for evaluating and benchmarking liveness detection methods could also facilitate fair comparisons and promote reproducibility across different studies. Lastly, advancements in hardware technologies, such as improved sensor designs and embedded processing capabilities, can play a vital role in enhancing the efficiency and real-world applicability of fingerprint liveness detection systems. Collaborative efforts between academia, industry, and regulatory bodies are essential to drive innovation and ensure the responsible development and deployment of liveness detection technologies in various domains, including cyber security, law enforcement, and mobile authentication. By addressing these research directives, the field of fingerprint liveness detection can continue to evolve and meet the growing demand for secure and reliable biometric authentication solutions.

Table 2: Comparative Analysis of Fingerprint Liveness Detection Methods

The Table 3 offers a comprehensive comparative analysis of various biometric cryptosystems, shedding light on different methods, datasets, metrics, accuracies, and limitations associated with these approaches. These biometric cryptosystems encompass a wide array of techniques such as fuzzy vault-based fingerprint cryptosystems, biometric key binding schemes, cryptographic authentication schemes based on discrete logarithm problems, and multi-biometric template security mechanisms, among others. Each method demonstrates specific strengths, including high accuracy, robust security, and effectiveness in protecting biometric templates. However, they also encounter challenges such as susceptibility to privacy attacks, computational complexity, limitations in dataset availability, and reliance on specific biometric modalities. Several future research directives emerge from this analysis. Firstly, there is a pressing need for the development of more efficient and secure biometric cryptosystems capable of addressing emerging security threats and vulnerabilities, including adversarial attacks, privacy breaches, and database-level attacks.  Secondly, research efforts should focus on enhancing the scalability and usability of biometric cryptosystems, particularly in real-world applications such as secure communication, access control, and identity verification. Thirdly, the exploration of novel cryptographic primitives, deep learning techniques, and hardware implementations could lead to innovative biometric cryptosystems with improved performance and reliability. Moreover, future research should prioritize the development of biometric cryptosystems that are compliant with regulatory requirements such as GDPR and HIPAA, while also ensuring user privacy and data protection. Additionally, efforts should be directed towards standardizing evaluation protocols and benchmarks to facilitate fair comparisons and reproducibility across different biometric cryptosystems. Furthermore, research should focus on the integration of biometric cryptosystems with emerging technologies such as blockchain and IoT to enhance security and interoperability in diverse application scenarios. By addressing these future research directives, the field of biometric cryptosystems can advance towards more secure, efficient, and user-centric solutions, thereby meeting the evolving needs of various domains including cyber security, healthcare, finance, and law enforcement.

Table 3: Comparative Analysis of Biometric Cryptosystems

The Table 4 presents a comparative analysis of cancellable biometric methods, offering insights into various techniques, datasets, metrics, accuracies, and limitations associated with these approaches. These methods encompass a wide range of strategies such as protection methods for fingerprint templates, generating masterprints for impersonation, one- factor cancellable biometric authentication schemes, secure triplet loss for training deep learning models, and multi- server authentication using cancellable biometrics and PUF, among others. Each method exhibits specific strengths, including high recognition accuracy, robustness against at- tacks, and effectiveness in protecting biometric templates. However, they also face challenges such as computational complexity, vulnerability to specific attacks, dependence on dataset characteristics, and limitations in adaptability to diverse biometric modalities. Several future research directions emerge from this analysis. Firstly, there is a need for the development of more efficient and scalable cancellable bio- metric techniques, particularly in addressing computational complexity issues and improving performance across di- verse biometric datasets. Secondly, research efforts should focus on enhancing the security and robustness of cancellable biometric methods against evolving threats and sophisticated attacks, including template inversion attacks, brute-force attacks, and adversarial manipulations. Thirdly, the exploration of novel approaches such as deep learning- based fusion techniques, chaotic-based cancellable systems, and image style transfer for biometric authentication could lead to innovative solutions with improved performance and reliability. Moreover, there is a need for standardized evaluation protocols and benchmarks to facilitate fair comparisons and reproducibility across different cancellable biometric methods. Additionally, future research should prioritize the development of cancellable biometric techniques that are user-friendly, privacy-preserving, and compliant with regulatory requirements such as GDPR and HIPAA. Further- more, efforts should be directed towards investigating the usability and acceptability of cancellable biometric systems in real-world applications such as access control, identity verification, and secure authentication. By addressing these future research directives, the field of cancellable biometric methods can advance towards more secure, efficient, and user-centric solutions, thereby meeting the evolving needs of various domains including cyber security, healthcare, finance, and law enforcement.

Table 4: Comparative Analysis of Cancellable Biometric Methods

The presented Table 5 offers a comprehensive analysis of various general-purpose user authentication methods, covering a wide range of techniques, datasets, metrics, accuracies, and limitations. These methods encompass diverse approaches such as orientation extraction, fingerprint reconstruction, fake biometric trait detection, local model-based classification, partial fingerprint matching, minutiae extraction evaluation, and multi-modal biometric authentication, among others. Each method demonstrates specific strengths, including high accuracy rates, robustness against attacks, and efficiency in authentication processes. However, they also encounter challenges such as dataset limitations, computational resource requirements, dependence on specific features, and vulnerabilities to spoofing attacks. Several future research directions can be identified from this analysis. Firstly, there is a need for the development of more robust and secure authentication methods, particularly in addressing vulnerabilities to spoofing attacks and enhancing resistance against adversarial manipulations. Secondly, research efforts should focus on improving the scalability and efficiency of authentication systems, especially in handling large-scale datasets and reducing computational resource requirements. Thirdly, the exploration of novel biometric features, fusion techniques, and machine learning algorithms could further enhance the accuracy and reliability of authentication methods across diverse modalities. Moreover, the development of privacy-preserving authentication techniques and the investigation of human-centric design principles could ensure the usability and acceptability of authentication systems in real-world scenarios. Lastly, efforts should be directed towards standardizing evaluation protocols and benchmarks to facilitate fair comparisons and reproducibility across different authentication methods. By addressing these future research directives, the field of general-purpose user authentication can advance towards more secure, efficient, and user-friendly authentication systems, thereby meeting the evolving needs of various applications such as cybersecurity, access control, and identity verification.

Table 5: Comparative Analysis of General Purpose User Authentication Methods

The Table 6 offers a comprehensive comparative analysis of various fingerprint indexing methods, encompassing different approaches, datasets, metrics, accuracies, and limitations. Notable methodologies include distributed frameworks, GPU-based systems, absolute registration approaches, minutiae-based algorithms, deep learning-based models, and privacy-preserving indexing techniques. These methods exhibit diverse strengths, such as high processing speeds, improved accuracies, and workload reduction, while also facing challenges like hardware/software dependencies, scalability issues, and computational costs. Several future research directions emerge from this analysis. Firstly, there’s a need for continued exploration of deep learning techniques in fingerprint indexing, especially in improving accuracy and reducing computational costs. Additionally, addressing scalability concerns and hardware dependencies would be crucial for real-world deployment. Secondly, the development of privacy-preserving indexing methods is vital to ensure the security and confidentiality of biometric data, especially in light of increasing concerns regarding data privacy. Thirdly, research efforts should focus on enhancing the robustness and generalization capabilities of indexing algorithms, particularly in handling diverse datasets with varying quality and characteristics. Moreover, investigating novel features and descriptors, as well as exploring multimodal approaches, could further improve indexing performance and reliability across different biometric modalities. Lastly, efforts should be directed towards standardizing evaluation protocols and benchmarks to facilitate fair comparisons and reproducibility across different indexing methods.

Table 6: Comparative Analysis of Fingerprint Indexing Methods

The Table 7 presents a comprehensive overview of various methodologies employed for infant and child biometric recognition, highlighting the diverse approaches, datasets, metrics, accuracies, and limitations associated with each method. Several studies focus on fingerprint recognition, utilizing techniques ranging from image enhancement and CNN-based ridge flow estimation to pre-trained CNN models. These approaches demonstrate promising accuracies, with [166] achieving up to 99.5% True Accept Rate (TAR) for infant fingerprints. However, challenges persist, such as the lack of automated alignment methods and issues with small fingerprint sizes and low image quality, as noted by multiple authors. Moreover, alternative biometric modalities like footprints, iris, and outer ear shape are explored, showing potential for high accuracy, as evidenced by [168] 99.30% accuracy in footprint identification and [170] low Equal Error Rate (EER) for outer ear shape recognition. Nonetheless, these methods encounter hurdles related to data collection challenges, limited dataset sizes, and transparency in processes and facilities. Future research in infant and child biometric recognition could address these limitations by focusing on several key areas. Firstly, efforts should be directed towards developing automated alignment techniques to enhance accuracy and efficiency, especially for infant fingerprints. Secondly, there is a need for larger and more diverse datasets to ensure robust model performance and generalization. Thirdly, exploration of multi-modal biometric systems, as suggested by [172], could provide enhanced accuracy and reliability by combining different biometric traits . Lastly, advancements in image acquisition technologies and computing resources could facilitate the collection of high-quality images and alleviate challenges associated with low image quality and computing resource requirements, as highlighted by [169]. By addressing these research directions, future studies can contribute towards the development of more robust, accurate, and reliable infant and child biometric recognition systems, thereby enhancing their applicability in various domains such as healthcare, security, and childcare.

Table 7: Comparative Analysis of Infant and Child Biometric Recognition Methods

prehensive analysis of the literature on latent fingerprint recognition and biometric identification systems. With a wide range of strategies being investigated, such as deep learning models, hybrid approaches, and sophisticated algo- rithmic techniques, these systems have undergone tremen- dous evolution. Due to this diversity, difficult fingerprint identification problems like partial prints, noise, and in- consistent quality have been significantly improved. The literature does, however, also highlight enduring difficul- ties. The comparison of various methods is made more difficult by the absence of standard evaluation methodolo- gies, and algorithmic bias and generalizability are raised by the small size and lack of diversity of the datasets that are currently accessible. Moreover, a lot of sophisticated techniques, especially those centred on deep learning, have a high computational complexity, which makes real-time implementation challenging. Future work in this area has to concentrate on creating uniform benchmarks, growing datasets, maximizing computational effectiveness, and re- solving moral issues with bias and privacy. Advantage, disadvantage, area of improvement and ethical implication of the fingerprint biometric studied in the manuscript are discussed below.

•   Advantage:

• Diverse Methodologies: The literature study demonstrates the wide variety of approaches, such as hybrid approaches, descriptor-based tactics, deep learning models, and different algorithmic advancements, that are employed in fingerprint recognition and biometric systems. Due to the variety of methods available, researchers are able to tackle several facets of latent fingerprint recognition issues, including handling partial prints, noise, and quality For instance, descriptor-based methods may provide computational efficiency, while deep learning models might enhance accuracy by discovering complex patterns in fingerprint data. The use of hybrid techniques, which blend conventional and cutting-edge methodologies, shows an effort to capitalize on each approach’s advantages and advance latent fingerprint recognition technology.
• Quantifiable Performance Metrics: Numerous studies provide comprehensive performance indicators, like F1 scores, recall, precision, and rank-1 accuracy, which are essential for evaluating the efficacy of various techniques. Benchmarking these measures against popular datasets such as FVC2004 and NIST SD27 is common practice. Researchers and practitioners find it easier to compare the effectiveness of different techniques, evaluate advances, and pinpoint areas that require ad- ditional development when such measurements are transparently reported. Because of its transparency, the community is able to monitor advancements and set performance standards for systems to come.
• Cutting-edge Techniques: Numerous novel methods have been presented, such as deep learning-based minutiae extraction, deformable minutiae clustering, and data augmentation using Generative Adversarial Networks (GAN). Especially for under-represented in- stances, GAN-based augmentation might artificially generate extra training data, hence improving the system’s resilience. Latent fingerprint identification systems are becoming more accurate and scalable thanks to deformable minutiae clustering and deep learning techniques. This makes them more suitable for use in forensic applications, particularly in challenging situa- tions where the prints are imperfect or deteriorated.

• Real-world Testing: The fact that numerous suggested techniques are evaluated on a range of databases and datasets, including NIST SD27 and FVC2004, is a note- worthy strength of the examined This makes it possible to test the algorithms in a variety of scenar- ios, simulating the unpredictability of the actual world. By doing this, scientists may evaluate how reliable and broadly applicable their methods are, making sure they work effectively outside of carefully regulated lab environments. For technology meant to be used in actual applications, including forensic investigations and security systems, this real-world applicability is essential.

•    Disadvantages:

• Lack of Uniformity in Evaluation Standards: Al- though a large number of research offer performance indicators, there is a notable deficiency in uniform assessment procedures between various Direct comparisons between approaches are challenging because to variations in benchmarking procedures, quality criteria, and dataset sizes. Because of this, some accuracy statistics might be inflated, which makes it more difficult to determine whether strategy is better in an unbiased manner. A technique that works well on a small, clean dataset, for instance, might not translate as well to larger, messier real-world data. Additionally, because there isn’t a single benchmark, reported accuracy gains may vary depending on the context and may not apply to different circumstances.
• Dataset Limitations: The small size and lack of diversity of datasets used for testing and training is a problem that keeps coming up in the literature. A lot of research papers admit that their datasets are too tiny or homogeneous to adequately capture the heterogeneity present in real-world circumstances, especially when it comes to latent print datasets. This suggests that the models may be over fitting to the unique features of these small datasets, which raises questions about their A system that works well on the dataset it was trained on but has trouble with under-represented populations or situations, including various skin kinds, lighting settings, or sensor types, may also be introduced by a lack of diversity.
• Computational Complexity: There is a drawback to several of the sophisticated methods, especially those that depend on deep learning: higher computational demands. These techniques frequently call for additional processing power, memory, and time, but they may provide considerable accuracy It is difficult to apply these techniques in real-time scenarios, such as live fingerprint recognition at security checkpoints or mobile devices with constrained computing power, due to the rise in computational complexity. For example, deep learning models can yield higher recognition rates, but their resource needs make them impractical for use in real-time, low-power devices.

•   Areas for Improvement:

• Standardized Benchmarks and Evaluation Protocols: The creation of uniform protocols for comparing finger- print recognition algorithms across various datasets is among the most urgent This would entail defining testing settings, dataset properties, and consistent evaluation criteria. Researchers may more precisely compare the performance of various approaches when a consistent framework is in place, which helps them determine which techniques are actually improving the state of the art. Exaggerated performance claims might be lessened by using a consistent evaluation process, which would guarantee that all approaches are examined under comparable circumstances [118].
• Larger and More Diverse Datasets: Future study should concentrate on increasing the amount and diversity of latent fingerprint databases in order to solve the limitations of the present investigations. This would include gathering information from a larger variety of sources and in a greater variety of settings, including various skin tones, ambient variables, and sensor kinds. By using larger, more representative datasets for algorithm training, researchers can improve system robustness and minimize bias. It would be easier to make sure that the algorithms can handle the whole range of real-world issues if more diverse latent prints were included, such as those with different levels of noise, distortion, and completeness [145].
• Improving Computational Efficiency: Because many of the methods are computationally demanding, optimization is required to shorten processing times without compromising For real-time applications, where speed is a crucial component, this is especially crucial. To lessen their processing footprint, researchers can concentrate on creating lightweight algorithms or improving already-existing deep learning systems. For example, the models could be deployed in resource- constrained contexts like mobile devices or edge computing platforms if they are optimized to perform well on GPUs or embedded systems [25].
• Integration with Other Biometric Modalities: Integrating fingerprint identification systems with other biometric modalities, such facial recognition or iris scanning, is another possible avenue for When one modality is impaired (a partial fingerprint, for example), this multi-modal method may enhance overall accuracy and resilience. Combining information from several biometric sources may also improve security by making it more difficult for hackers to impersonate the system. Future developments may proceed in the direction of sensor fusion technique research and the creation of reliable multi-modal biometric authentication algorithms [42].

• Ethical Considerations and Privacy: Future research must also address ethical issues including privacy concerns, data protection, and algorithmic bias given the growing usage of biometric technologies in sensitive applications like security and forensics. Strong legal frameworks are especially necessary for forensic applications in order to guarantee the validity and acceptability of the data gathered using these systems. Researchers should investigate the moral ramifications of gathering and using biometric data, making sure privacy laws like GDPR are followed, and creating clear procedures for handling user consent and data [180].

•   Ethical Implications:

• Privacy Concerns: Fingerprint biometric systems create serious privacy concerns since they rely on the gathering, storing, and processing of private information. Biometric information, like fingerprints, is unchangeable after it is compromised, unlike passwords or other identity-related Biometric data is particularly susceptible to exploitation because of this. Identity theft, unapproved tracking, or monitoring might result from unauthorized access to fingerprint databases or data breaches. These worries are heightened by the collecting of biometric data, frequently without users’ express or informed agreement. The collection, storage, and sharing of this data must adhere to strict processes in order to comply with data protection legislation, such as the General Data Protection Regulation (GDPR) in Europe. Crucial to ethical compliance are letting users know how their biometric data is being used and providing them with the choice to opt out. To reduce these concerns, biometric systems should place a high priority on openness, safe encryption, and privacy- preserving methods like homomorphic encryption or cancelable biometrics.
• Algorithmic Biases: The possibility of algorithmic bias in fingerprint biometric technologies presents another ethical dilemma. Research has indicated that the accuracy of biometric systems can vary based on the demographic attributes of the people being scanned, including age, gender, and skin tone. For instance, under representation of particular populations in training datasets may result in reduced accuracy for those groups. Particularly in forensic or security scenarios, where a false match or inability to identify a legitimate individual can have dire effects, this bias can have catastrophic Misidentification in forensic investigations may result in false allegations, and in security applications, specific populations may be unjustly singled out or denied access. In order to solve this problem, more diverse datasets must be used, and algorithms must be developed with fairness standards in place. These problems can be lessened by routinely reviewing biometric systems for bias and including a variety of demographic groups in the testing process.

• Legal Considerations: Particularly in forensic and government use cases, fingerprint biometric technologies have a wide range of legal The reliability and admissibility of biometric evidence in court is one of the main issues. Strict legal requirements must be met by forensic fingerprint identification systems in order for the evidence they produce to be trustworthy and supported by science. But many contemporary biometric algorithms are “black-box,” especially those that rely on deep learning, which makes it challenging to under- stand how they arrive at a given judgement. This raises concerns about the accountability and transparency of these systems. Furthermore, there are jurisdictional differences in the legality of biometric data collecting, thus it is crucial to make sure local regulations are followed. While some nations have more permissive rules, others have tight legislation requiring express authorization for the use of biometric data. The ethical application of biometric technology must be covered by legal frameworks, with an emphasis on data protection, user permission, and making sure that systems are operated in a way that respects people’s rights.

This review study on fingerprint biometrics has some major limitations, one of them being its scant discussion of recent developments in fingerprint sensing technology and their potential applications. Although the study thoroughly discusses the current state of the art in fingerprint recognition, it does not go into great detail into cutting edge advances like new capture methods, sophisticated sensing materials, or the effects of recent advancements in fingerprint sensor downsizing. The absence of these developments implies that the evaluation may not accurately reflect the state of the field today and in the future, given the potential impact they could have on the precision, usability, and integration of fingerprint biometrics in a variety of scenarios. This gap may allow significant technical trends to go unnoticed, along with the implications they have for research and real-world applications.

8. Conclusion

The adoption of creative solutions to current problems is essential to the future viability of biometric authentication. This entails creating reliable, scalable algorithms that can efficiently handle a variety of biometric data. It is imperative to establish uniform standards and evaluation procedures to guarantee impartial assessments of diverse authentication techniques. Strict ethical guidelines and openness campaigns are necessary to solve issues with algorithmic bias and data privacy in order to win over the public’s trust. Multimodal fusion techniques that integrate various biometric features can greatly improve dependability. Furthermore, maximizing hardware technological developments is critical to increasing accuracy and productivity. The goal of research should be to develop biometric systems that are easy to use and prioritize both regulatory compliance and privacy protection. Collaboration between academic re- searchers, business personnel, and regulatory organizations is essential to spur innovation and guarantee the responsible application of biometric technologies across various sectors. By implementing these tactics, biometric authentication will be able to advance, satisfy the increasing need for reliable and secure authentication methods, and improve security and user experience across a range of applications.

Conflict of Interest The authors declare no conflict of interest.

1. R. Ramotowski, Lee and Gaensslen’s advances in fingerprint technology, CRC press, 2012.
2. Y. Xu, G. Lu, Y. Lu, D. Zhang, “High resolution fingerprint recog- nition using pore and edge descriptors”, Pattern Recognition Letters, vol. 125, pp. 773–779, 2019, doi:10.1016/j.patrec.2019.08.006.
3. P. Tertychnyi, C. Ozcinar, G. Anbarjafari, “Low-quality fingerprint classification using deep neural network”, IET Biometrics, vol. 7, no. 6, pp. 550–556, 2018, doi:10.1049/iet-bmt.2018.5074.
4. A. A. Paulino, J. Feng, A. K. Jain, “Latent fingerprint matching using descriptor-based hough transform”, IEEE Transactions on Information Forensics and Security, vol. 8, no. 1, pp. 31–45, 2012, doi:10.1109/ijcb.2011.6117483.
5. S. S. Arora, E. Liu, K. Cao, A. K. Jain, “Latent fingerprint matching: performance gain via feedback from exemplar prints”, IEEE Transac- tions on Pattern Analysis and Machine Intelligence, vol. 36, no. 12, pp. 2452–2465, 2014, doi:10.1109/tpami.2014.2330609.
6. M. A. Medina-Pérez, A. M. Moreno, M. Á. F. Ballester, M. García- Borroto, O. Loyola-González, L. Altamirano-Robles, “Latent finger- print identification using deformable minutiae clustering”, Neurocom- puting, vol. 175, pp. 851–865, 2016, doi:10.1016/j.neucom.2015.05.130.
7. A. Sankaran, A. Jain, T. Vashisth, M. Vatsa, R. Singh, “Adaptive latent fingerprint segmentation using feature selection and random decision forest classification”, Information Fusion, vol. 34, pp. 1–15, 2017, doi:10.1016/j.inffus.2016.05.002.
8. W. Bian, S. Ding, W. Jia, “Collaborative filtering model for enhancing fingerprint image”, IET Image Processing, vol. 12, no. 1, pp. 149–157, 2018, doi:10.1049/iet-ipr.2017.0059.
9. J. Li, J. Feng, C.-C. J. Kuo, “Deep convolutional neural network for latent fingerprint enhancement”, Signal Processing: Image Communi- cation, vol. 60, pp. 52–63, 2018, doi:10.1016/j.image.2017.08.010.
10. K. Cao, A. K. Jain, “Automated latent fingerprint recognition”, IEEE transactions on pattern analysis and machine intelligence, vol. 41, no. 4, pp. 788–800, 2018, doi:10.1109/tpami.2018.2818162.
11. R. P. Krish, J. Fierrez, D. Ramos, F. Alonso-Fernandez, J. Bigun, “Improving automated latent fingerprint identification using ex- tended minutia types”, Information Fusion, vol. 50, pp. 9–19, 2019, doi:10.1016/j.inffus.2018.10.001.
12. K. Cao, D.-L. Nguyen, C. Tymoszek, A. K. Jain, “End-to-end latent fingerprint search”, IEEE Transactions on Information Forensics and Security, vol. 15, pp. 880–894, 2019, doi:10.1109/tifs.2019.2930487.
13. A. J. Sanchez-Fernandez, L. F. Romero, D. Peralta, M. A. Medina- Pérez, Y. Saeys, F. Herrera, S. Tabik, “Asynchronous process- ing for latent fingerprint identification on heterogeneous cpu- gpu systems”, IEEE Access, vol. 8, pp. 124236–124253, 2020, doi:10.1109/access.2020.3005476.
14. N. D. Kalka, M. Beachler, R. A. Hicklin, “Lqmetric: a latent finger- print quality metric for predicting afis performance and assessing the value of latent fingerprints”, Journal of Forensic Identification, vol. 70, no. 4, pp. 443–463, 2020.
15. X. Huang, P. Qian, M. Liu, “Latent fingerprint image enhance- ment based on progressive generative adversarial network”, “Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition Workshops”, pp. 800–801, 2020, doi:10.1109/cvprw50498.2020.00408.
16. U. U. Deshpande, V. Malemath, S. M. Patil, S. V. Chaugule, “End- to-end automated latent fingerprint identification with improved dcnn-fft enhancement”, Frontiers in Robotics and AI, vol. 7, p. 594412, 2020, doi:10.3389/frobt.2020.594412.
17. M. Liu, P. Qian, “Automatic segmentation and enhancement of latent fingerprints using deep nested unets”, IEEE Transactions on Information Forensics and Security, vol. 16, pp. 1709–1719, 2020, doi:10.1109/tifs.2020.3039058.
18. S. Gu, J. Feng, J. Lu, J. Zhou, “Latent fingerprint registration via matching densely sampled points”, IEEE Transactions on In- formation Forensics and Security, vol. 16, pp. 1231–1244, 2020, doi:10.1109/tifs.2020.3032041.
19. D. Agarwal, A. Bansal, “A utility of pores as level 3 features in latent fingerprint identification”, Multimedia Tools and Applications, vol. 80, no. 15, pp. 23605–23624, 2021, doi:10.1007/s11042-020-10207-x.
20. U. U. Deshpande, V. Malemath, S. M. Patil, S. V. Chaugule, “Latent fingerprint identification system based on a local combination of minutiae feature points”, SN Computer Science, vol. 2, no. 3, p. 206, 2021, doi:10.1007/s42979-021-00615-7.
21. S. M. Hilles, A. Liban, O. A. Miaikil, A. M. Altrad, Y. A. B. El- Ebiary, M. M. Hilles, J. Contreras, “Latent fingerprint enhancement and segmentation technique based on hybrid edge adaptive dtv model”, “2021 2nd International Conference on Smart Comput- ing and Electronic Enterprise (ICSCEE)”, pp. 8–13, IEEE, 2021, doi:10.1109/icscee50312.2021.9498025.
22. S. M. Hilles, A. Liban, A. M. Altrad, O. A. Miaikil, Y. A. B. El- Ebiary, J. Contreras, M. M. Hilles, “Adaptive latent fingerprint image segmentation and matching using chan-vese technique based on edtv model”, “2021 2nd International Conference on Smart Com- puting and Electronic Enterprise (ICSCEE)”, pp. 2–7, IEEE, 2021, doi:10.1109/icscee50312.2021.9497996.
23. U. U. Deshpande, V. Malemath, S. M. Patil, S. V. Chaugule, “Auto- matic latent fingerprint identification system using scale and rota- tion invariant minutiae features”, International Journal of Information Technology, vol. 14, no. 2, pp. 1025–1039, 2022, doi:10.1007/s41870- 020-00508-7.
24. H. İ. Öztürk, B. Selbes, Y. Artan, “Minnet: Minutia patch embedding network for automated latent fingerprint recognition”, “Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recog- nition”, pp. 1627–1635, 2022, doi:10.1109/cvprw56347.2022.00169.
25. S. Gu, J. Feng, J. Lu, J. Zhou, “Latent fingerprint indexing: Ro- bust representation and adaptive candidate list”, IEEE Transactions on Information Forensics and Security, vol. 17, pp. 908–923, 2022, doi:10.1109/tifs.2022.3154296.
26. N. D. S. Cunha, H. M. Gomes, L. V. Batista, “Residual m-net with frequency-domain loss function for latent fingerprint en- hancement”, “2022 35th SIBGRAPI Conference on Graphics, Pat- terns and Images (SIBGRAPI)”, vol. 1, pp. 198–203, IEEE, 2022, doi:10.1109/sibgrapi55357.2022.9991793.
27. E. Marasco, M. He, L. Tang, Y. Tao, “Demographic effects in latent fingerprint matching and their relation to image quality”, “Proceed- ings of the 2022 7th International Conference on Machine Learning Technologies”, pp. 170–179, 2022, doi:10.1145/3529399.3529427.
28. A. B. V. Wyzykowski, A. K. Jain, “Synthetic latent finger- print generator”, “Proceedings of the IEEE/CVF Winter Confer- ence on Applications of Computer Vision”, pp. 971–980, 2023, doi:10.1109/wacv56688.2023.00103.
29. Y. Zhu, X. Yin, J. Hu, “Fingergan: a constrained fingerprint generation scheme for latent fingerprint enhancement”, IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, doi:10.1109/tpami.2023.3236876.
30. S. A. Grosz, A. K. Jain, “Latent fingerprint recognition: Fusion of local and global embeddings”, IEEE Transactions on Information Forensics and Security, 2023, doi:10.1109/tifs.2023.3314207.
31. R. Jindal, S. Singla, “Latent fingerprint recognition using hybrid ant colony optimization and cuckoo search.”, Int. Arab J. Inf. Technol., vol. 20, no. 1, pp. 19–28, 2023, doi:10.34028/iajit/20/1/3.
32. A. B. V. Wyzykowski, A. K. Jain, “A universal latent fingerprint enhancer using transformers”, arXiv preprint arXiv:2306.00231, 2023, doi:10.48550/arXiv.2306.00231.
33. R. K. Dubey, J. Goh, V. L. Thing, “Fingerprint liveness detection from single image using low-level features and shape analysis”, IEEE Transactions on Information Forensics and Security, vol. 11, no. 7, pp. 1461–1475, 2016, doi:10.1109/tifs.2016.2535899.
34. Z. Xia, R. Lv, Y. Zhu, P. Ji, H. Sun, Y.-Q. Shi, “Fingerprint liveness de- tection using gradient-based texture features”, Signal, Image and Video Processing, vol. 11, pp. 381–388, 2017, doi:10.1007/s11760-016-0936-z.
35. T. Chugh, K. Cao, A. K. Jain, “Fingerprint spoof buster: Use of minutiae-centered patches”, IEEE Transactions on Informa- tion Forensics and Security, vol. 13, no. 9, pp. 2190–2202, 2018, doi:10.1109/tifs.2018.2812193.
36. C. Yuan, X. Sun, Q. J. Wu, “Difference co-occurrence matrix using bp neural network for fingerprint liveness detection”, Soft Computing, vol. 23, no. 13, pp. 5157–5169, 2019, doi:10.1007/s00500-018-3182-1.
37. C. Yuan, X. Chen, P. Yu, R. Meng, W. Cheng, Q. J. Wu, X. Sun, “Semi- supervised stacked autoencoder-based deep hierarchical semantic feature for real-time fingerprint liveness detection”, Journal of Real- Time Image Processing, vol. 17, pp. 55–71, 2020, doi:10.1007/s11554- 019-00928-0.
38. J. Fei, Z. Xia, P. Yu, F. Xiao, “Adversarial attacks on fingerprint liveness detection”, EURASIP Journal on Image and Video Processing, vol. 2020, pp. 1–11, 2020, doi:10.1186/s13640-020-0490-z.
39. Y. Zhang, C. Gao, S. Pan, Z. Li, Y. Xu, H. Qiu, “A score- level fusion of fingerprint matching with fingerprint live- ness detection”, IEEE Access, vol. 8, pp. 183391–183400, 2020, doi:10.1109/access.2020.3027846.
40. W. Jian, Y. Zhou, H. Liu, “Densely connected convolutional network optimized by genetic algorithm for fingerprint live- ness detection”, IEEE Access, vol. 9, pp. 2229–2243, 2020, doi:10.1109/access.2020.3047723.
41. O. F. Onifade, P. Akinde, F. O. Isinkaye, “Circular gabor wavelet algorithm for fingerprint liveness detection”, Journal of Advanced Computer Science & Technology, vol. 9, no. 1, pp. 1–5, 2020, doi:10.14419/jacst.v9i1.29908.
42. R. Agarwal, A. S. Jalal, K. Arya, “A multimodal liveness detection using statistical texture features and spatial analysis”, Multime- dia Tools and Applications, vol. 79, no. 19, pp. 13621–13645, 2020, doi:10.1007/s11042-019-08313-6.
43. Y. Zhang, S. Pan, X. Zhan, Z. Li, M. Gao, C. Gao, “Fldnet: Light dense cnn for fingerprint liveness detection”, IEEE Access, vol. 8, pp. 84141–84152, 2020, doi:10.1109/access.2020.2990909.
44. J. Kolberg, M. Grimmer, M. Gomez-Barrero, C. Busch, “Anomaly detection with convolutional autoencoders for fingerprint pre- sentation attack detection”, IEEE Transactions on Biometrics, Be- havior, and Identity Science, vol. 3, no. 2, pp. 190–202, 2021, doi:10.1109/tbiom.2021.3050036.
45. C. Yuan, S. Jiao, X. Sun, Q. J. Wu, “Mfffld: A multimodal-feature- fusion-based fingerprint liveness detection”, IEEE Transactions on Cognitive and Developmental Systems, vol. 14, no. 2, pp. 648–661, 2021, doi:10.1109/tcds.2021.3062624.
46. S. Agarwal, C. R. Chowdary, V. Sourabh, “Eazy learning: An adap- tive variant of ensemble learning for fingerprint liveness detection”, arXiv preprint arXiv:2103.02207, 2021, doi:10.48550/arXiv.2103.02207.
47. S. B. Sandouka, Y. Bazi, N. Alajlan, “Transformers and generative adversarial networks for liveness detection in multitarget fingerprint sensors”, Sensors, vol. 21, no. 3, p. 699, 2021, doi:10.3390/s21030699.
48. A. Verma, V. K. Gupta, S. Goel, A. K. Yadav, D. Yadav, et al., “Mod- eling fingerprint presentation attack detection through transient liveness factor-a person specific approach.”, Traitement du Signal, vol. 38, no. 2, 2021, doi:10.18280/ts.380206.
49. C. Yuan, Q. Cui, X. Sun, Q. J. Wu, S. Wu, “Fingerprint liveness detection using an improved cnn with the spatial pyramid pooling structure”, “Advances in Computers”, vol. 120, pp. 157–193, Elsevier, 2021, doi:10.1016/bs.adcom.2020.10.002.
50. Z. İ. Özkiper, Z. Turgut, T. Atmaca, M. A. Aydın, “Fingerprint liveness detection using deep learning”, “2022 9th International Conference on Future Internet of Things and Cloud (FiCloud)”, pp. 129–135, IEEE, 2022, doi:10.1109/ficloud57274.2022.00025.
51. R. C. Contreras, L. G. Nonato, M. Boaventura, I. A. G. Boaventura, F. L. Dos Santos, R. B. Zanin, M. S. Viana, “A new multi-filter framework for texture image representation improvement using set of pattern descriptors to fingerprint liveness detection”, IEEE Access, vol. 10, pp. 117681–117706, 2022, doi:10.1109/access.2022.3218335.
52. A. Almehmadi, “A behavioral-based fingerprint liveness and will- ingness detection system”, Applied Sciences, vol. 12, no. 22, p. 11460, 2022, doi:10.3390/app122211460.
53. S. Agarwal, A. Rattani, C. R. Chowdary, “A-ilearn: An adap- tive incremental learning model for spoof fingerprint detec- tion”, Machine Learning with Applications, vol. 7, p. 100210, 2022, doi:10.1016/j.mlwa.2021.100210.
54. D. Agarwal, A. Bansal, “Fingerprint liveness detection through fusion of pores perspiration and texture features”, Journal of King Saud University-Computer and Information Sciences, vol. 34, no. 7, pp. 4089–4098, 2022, doi:10.1016/j.jksuci.2020.10.003.
55. K. Zhang, S. Huang, E. Liu, H. Zhao, “Lfldnet: Lightweight finger- print liveness detection based on resnet and transformer”, Sensors, vol. 23, no. 15, p. 6854, 2023, doi:10.3390/s23156854.
56. A. Galli, M. Gravina, S. Marrone, D. Mattiello, C. Sansone, “Ad- versarial liveness detector: Leveraging adversarial perturbations in fingerprint liveness detection”, IET Biometrics, vol. 12, no. 2, pp. 102–111, 2023, doi:10.1049/bme2.12106.
57. M. Nishanth, H. K. MR, K. N. MG, S. Kamal, T. Rao, K. Ashwini, “Fingerprint liveness detection using deep learning”, “2023 Interna- tional Conference on Computational Intelligence for Information, Security and Communication Applications (CIISCA)”, pp. 383–388, IEEE, 2023, doi:10.1109/ciisca59740.2023.00079.
58. Y. Myshkovskyi, M. Nazarkevych, “Robustness of fingerprint live- ness detection based on convolutional neural networks”, https://ceur- ws.org/Vol-3550/short13.pdf, 2023.
59. C. Li, J. Hu, “A security-enhanced alignment-free fuzzy vault-based fingerprint cryptosystem using pair-polar minutiae structures”, IEEE Transactions on Information Forensics and Security, vol. 11, no. 3, pp. 543–555, 2015, doi:10.1109/tifs.2015.2505630.
60. T. Murakami, T. Ohki, K. Takahashi, “Optimal sequential fusion for multibiometric cryptosystems”, Information fusion, vol. 32, pp. 93–108, 2016, doi:10.1016/j.inffus.2016.02.002.
61. Z. Jin, A. B. J. Teoh, B.-M. Goi, Y.-H. Tay, “Biometric cryptosystems: a new biometric key binding and its implementation for finger- print minutiae-based representation”, Pattern Recognition, vol. 56, pp. 50–62, 2016, doi:10.1016/j.patcog.2016.02.024.
62. S. Rajendran, M. Doraipandian, “Biometric template security trig- gered by two dimensional logistic sine map”, Procedia computer science, vol. 143, pp. 794–803, 2018, doi:10.1016/j.procs.2018.10.387.
63. A. A. Al-Saggaf, “Secure method for combining cryptography with iris biometrics.”, J. Univers. Comput. Sci., vol. 24, no. 4, pp. 341–356, 2018.
64. G. Panchal, D. Samanta, S. Barman, “Biometric-based cryptog- raphy for digital content protection without any key storage”, Multimedia Tools and Applications, vol. 78, pp. 26979–27000, 2019, doi:10.1007/s11042-017-4528-x.
65. A. A. Khan, V. Kumar, M. Ahmad, “An elliptic curve cryptography based mutual authentication scheme for smart grid communications using biometric approach”, Journal of King Saud University-Computer and Information Sciences, 2019, doi:10.1016/j.jksuci.2019.04.013.
66. S. Sapkal, S. Kakarwal, R. Deshmukh, “Template security of mul- timodal biometric system with face and fingerprint images using fuzzy vault method”, CSI Journal of Computing, vol. 3, no. 3, pp. 36–40, 2020, doi:10.21275/v5i3.nov162389.
67. V. Rajasekar, J. Premalatha, K. Sathya, “Enhanced biometric recog- nition for secure authentication using iris preprocessing and hy- perelliptic curve cryptography”, Wireless communications and mobile computing, vol. 2020, pp. 1–15, 2020, doi:10.21203/rs.2.23196/v1.
68. F. S. Babamir, M. Kırcı, “A multibiometric cryptosystem for user authentication in client-server networks”, Computer Networks, vol. 181, p. 107427, 2020, doi:10.1016/j.comnet.2020.107427.
69. R. A. Rajan, P. Kumaran, “Multi-biometric cryptosystem using graph for secure cloud authentication”, Journal of Intelligent & Fuzzy Systems, vol. 38, no. 5, pp. 6437–6444, 2020, doi:10.3233/jifs-179724.
70. R. Dwivedi, S. Dey, M. A. Sharma, A. Goel, “A fingerprint based crypto-biometric system for secure communication”, Journal of Am- bient Intelligence and Humanized Computing, vol. 11, pp. 1495–1509, 2020, doi:10.1007/s12652-019-01437-5.
71. P. S. Chanukya, T. Thivakaran, “Multimodal biometric cryptosystem for human authentication using fingerprint and ear”, Multimedia Tools and Applications, vol. 79, no. 1, pp. 659–673, 2020, doi:10.1007/s11042- 019-08123-w.
72. J. Peng, B. Yang, B. B. Gupta, A. A. Abd El-Latif, “A biometric cryp- tosystem scheme based on random projection and neural network”, Soft Computing, vol. 25, pp. 7657–7670, 2021, doi:10.1007/s00500-021- 05732-2.
73. S. Barzut, M. Milosavljević, S. Adamović, M. Saračević, N. Maček,
    M. Gnjatović, “A novel fingerprint biometric cryptosystem based on convolutional neural networks”, Mathematics, vol. 9, no. 7, p. 730, 2021, doi:10.3390/math9070730.
74. W. El-Shafai, F. A. H. E. Mohamed, H. M. Elkamchouchi,
    M. Abd-Elnaby, A. Elshafee, “Efficient and secure cancelable biometric authentication framework based on genetic encryp- tion algorithm”, IEEE Access, vol. 9, pp. 77675–77692, 2021, doi:10.1109/access.2021.3082940.
75. H. A. A. El-Hameed, N. Ramadan, W. El-Shafai, A. A. Khalaf, H. E. H. Ahmed, S. E. Elkhamy, F. E. A. El-Samie, “Cancelable biometric secu- rity system based on advanced chaotic maps”, The Visual Computer, pp. 1–17, 2021, doi:10.1007/s00371-021-02276-2.
76. O. Ouda, K. Nandakumar, A. Ross, “Cancelable biometrics vault: A secure key-binding biometric cryptosystem based on chaffing and winnowing”, “2020 25th International Conference on Pattern Recognition (ICPR)”, pp. 8735–8742, IEEE, 2021, doi:10.1109/icpr48806.2021.9412957.
77. R. Sreemol, M. S. Kumar, A. Sreekumar, “Improvement of security in multi-biometric cryptosystem by modulus fuzzy vault algorithm”, “2021 International conference on advances in computing and com- munications (ICACC)”, pp. 1–7, IEEE, 2021, doi:10.1109/icacc- 202152719.2021.9708136.
78. L. A. Elrefaei, A. M. Al-Mohammadi, “Machine vision gait-based biometric cryptosystem using a fuzzy commitment scheme”, Journal of King Saud University-Computer and Information Sciences, vol. 34, no. 2, pp. 204–217, 2022, doi:10.1016/j.jksuci.2019.10.011.
79. X. Chang, W. Li, A. Yan, P. W. M. Tsang, T.-C. Poon, “Asymmetric cryptosystem based on optical scanning cryptography and elliptic curve algorithm”, Scientific Reports, vol. 12, no. 1, p. 7722, 2022, doi:10.21203/rs.3.rs-1148931/v1.
80. A. Kuznetsov, D. Zakharov, E. Frontoni, L. Romeo, R. Rosati, “Deep learning based fuzzy extractor for generating strong keys from biometric face images”, “2022 IEEE 9th Interna- tional Conference on Problems of Infocommunications, Sci- ence and Technology (PIC S&T)”, pp. 421–426, IEEE, 2022, doi:10.1109/picst57299.2022.10238643.
81. N. Hamian, M. Bayat, M. R. Alaghband, Z. Hatefi, S. M. Pournaghi, “Blockchain-based user re-enrollment for biometric authentication systems”, International Journal of Electronics and Information Engineer- ing, vol. 14, no. 1, pp. 18–38, 2022, doi:10.6636/IJEIE.202206.
82. S. D. Patil, R. Raut, R. H. Jhaveri, T. A. Ahanger, P. V. Dhade, A. B. Kathole, K. N. Vhatkar, et al., “Robust authentication system with privacy preservation of biometrics”, Security and Communication Networks, vol. 2022, 2022, doi:10.1155/2022/7857975.
83. L. A. Abou Elazm, W. El-Shafai, S. Ibrahim, M. G. Egila, H. Shawkey, M. K. Elsaid, N. F. Soliman, H. N. AlEisa, F. E. Abd El-Samie, “Efficient hardware design of a secure cancellable biometric cryp- tosystem”, Intell. Autom. Soft Comput., vol. 36, no. 1, pp. 929–955, 2023, doi:10.32604/iasc.2023.031386.
84. H. A. A. Eldawy, W. El-Shafai, E. E.-D. Hemdan, G. M. El-Banby, F. E. A. El-Samie, “A robust cancellable face and palmprint recogni- tion system based on 3d optical chaos-dna cryptosystem”, Optical and Quantum Electronics, vol. 55, no. 11, p. 970, 2023, doi:10.1007/s11082- 023-04840-7.
85. P. Kaur, N. Kumar, “Enhanced biometric cryptosystem using ear & iris modality based on binary robust independent elementary feature”, “2023 6th International Conference on Information Sys- tems and Computer Networks (ISCON)”, pp. 1–6, IEEE, 2023, doi:10.1109/iscon57294.2023.10112197.
86. A. Sedik, A. A. A. El-Latif, M. A. Wani, F. E. A. El-Samie, N. A.-S. Bauomy, F. G. Hashad, “Efficient multi-biometric secure-storage scheme based on deep learning and crypto-mapping techniques”, Mathematics, vol. 11, no. 3, p. 703, 2023, doi:10.3390/math11030703.
87. S. Nagaraju, R. Nagendra, S. Balasundaram, R. K. Kumar, “Bio- metric key generation and multi round aes crypto system for im- proved security”, Measurement: Sensors, vol. 30, p. 100931, 2023, doi:10.1016/j.measen.2023.100931.
88. M. Sandhya, M. V. Prasad, “Securing fingerprint templates using fused structures”, IET Biometrics, vol. 6, no. 3, pp. 173–182, 2017, doi:10.1049/iet-bmt.2016.0008.
89. A. Roy, N. Memon, A. Ross, “Masterprint: Exploring the vulner- ability of partial fingerprint-based authentication systems”, IEEE Transactions on Information Forensics and Security, vol. 12, no. 9, pp. 2013–2025, 2017, doi:10.1109/tifs.2017.2691658.
90. J. Kim, A. B. J. Teoh, “One-factor cancellable biometrics based on indexing-first-order hashing for fingerprint authentication”, “2018 24th International Conference on Pattern Recognition (ICPR)”, pp. 3108–3113, IEEE, 2018, doi:10.1109/icpr.2018.8545565.
91. X. Wang, H. Li, “One-factor cancellable palmprint recognition scheme based on oiom and minimum signature hash”, IEEE Access, vol. 7, pp. 131338–131354, 2019, doi:10.1109/access.2019.2938019.
92. G. S. Walia, K. Aggarwal, K. Singh, K. Singh, “Design and analysis of adaptive graph-based cancelable multi-biometrics approach”, IEEE Transactions on Dependable and Secure Computing, vol. 19, no. 1, pp. 54–66, 2020, doi:10.1109/tdsc.2020.2997558.
93. J. R. Pinto, M. V. Correia, J. S. Cardoso, “Secure triplet loss: Achieving cancelability and non-linkability in end-to-end deep biometrics”, IEEE Transactions on Biometrics, Behavior, and Identity Science, vol. 3, no. 2, pp. 180–189, 2020, doi:10.1109/tbiom.2020.3046620.
94. H. Wang, X. Dong, Z. Jin, A. B. J. Teoh, M. Tistarelli, “Security analysis of cancellable biometrics using constrained-optimized similarity-based attack”, arXiv preprint arXiv:2006.13051, 2020, doi:10.1109/wacvw52041.2021.00012.
95. A. K. Trivedi, D. M. Thounaojam, S. Pal, “Non-invertible cancellable fingerprint template for fingerprint biometric”, Computers & Security, vol. 90, p. 101690, 2020, doi:10.1016/j.cose.2019.101690.
96. U. Sharma, P. Tomar, S. S. Ali, N. Saxena, R. S. Bhadoria, “Optimized authentication system with high security and privacy”, Electronics, vol. 10, no. 4, p. 458, 2021, doi:10.3390/electronics10040458.
97. H. Wang, X. Dong, Z. Jin, A. B. J. Teoh, M. Tistarelli, “Interpretable se- curity analysis of cancellable biometrics using constrained-optimized similarity-based attack”, “Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision”, pp. 70–77, 2021, doi:10.1109/wacvw52041.2021.00012.
98. N. F. Soliman, A. D. Algarni, W. El-Shafai, F. E. A. El-Samie, G. Banby, et al., “An efficient gcd-based cancelable biometric algorithm for single and multiple biometrics.”, Computers, Materials & Continua, vol. 69, no. 2, 2021, doi:10.32604/cmc.2021.016980.
99. W. Yang, S. Wang, M. Shahzad, W. Zhou, “A cancelable biometric au- thentication system based on feature-adaptive random projection”, Journal of Information Security and Applications, vol. 58, p. 102704, 2021, doi:10.1016/j.jisa.2020.102704.
100. H. Zhang, W. Bian, B. Jie, D. Xu, J. Zhao, “A complete user au- thentication and key agreement scheme using cancelable biomet- rics and puf in multi-server environment”, IEEE Transactions on Information Forensics and Security, vol. 16, pp. 5413–5428, 2021, doi:10.1109/tifs.2021.3128826.
101. X. Dong, Z. Jin, L. Zhao, Z. Guo, “Biocancrypto: An ldpc coded bio-cryptosystem on fingerprint cancellable template”, “2021 IEEE international joint conference on biometrics (IJCB)”, pp. 1–8, IEEE, 2021, doi:10.1109/ijcb52358.2021.9484391.
102. B. Samira, R. H. Lamia, E. B. A. Najoua, “Biometric template security using watermarking reinforcement based cancellable transforma- tion”, “2021 International Conference on Cyberworlds (CW)”, pp. 270–277, IEEE, 2021, doi:10.1109/cw52790.2021.00052.
103. T. M. Dang, T. D. Nguyen, T. Hoang, H. Kim, A. B. J. Teoh, D. Choi, “Avet: a novel transform function to improve cancellable biometrics security”, IEEE Transactions on Information Forensics and Security, vol. 18, pp. 758–772, 2022, doi:10.1109/tifs.2022.3230212.
104. B. A. El-Rahiem, M. Amin, A. Sedik, F. E. A. E. Samie, A. M. Iliyasu, “An efficient multi-biometric cancellable biometric scheme based on deep fusion and deep dream”, Journal of Ambient Intelligence and Hu- manized Computing, pp. 1–13, 2022, doi:10.1007/s12652-021-03513-1.
105. A. M. Ayoup, A. A. Khalaf, W. El-Shafai, F. E. A. El-Samie, F. Alrad- dady, S. M. S. Eldin, “Cancellable multi-biometric template genera- tion based on arnold cat map and aliasing.”, Computers, Materials & Continua, vol. 72, no. 2, 2022, doi:10.32604/cmc.2022.025902.
106. J. Kim, Y. G. Jung, A. B. J. Teoh, “Multimodal biometric template pro- tection based on a cancelable softmaxout fusion network”, Applied Sciences, vol. 12, no. 4, p. 2023, 2022, doi:10.3390/app12042023.
107. A. M. Ayoup, A. Khalaf, F. Alraddady, F. Abd El-Samie, W. El- Safai, S. Eldin, “Selective cancellable multi-biometric template generation scheme based on multi-exposure feature fusion”, In- tell. Autom. Soft Comput., vol. 33, no. 1, pp. 549–565, 2022, doi:10.32604/iasc.2022.024379.
108. A. Sedik, A. A. A. El-Latif, M. El-Affendi, H. Mostafa, “A cancelable biometric system based on deep style transfer and symmetry check for double-phase user authentication”, Symmetry, vol. 15, no. 7, p. 1426, 2023, doi:10.3390/sym15071426.
109. S. M. S. Eldin, A. Sedik, S. S. Alshamrani, A. M. Ayoup, “Cancellable multi-biometric feature veins template generation based on sha- 3 hashing.”, Computers, Materials & Continua, vol. 75, no. 1, 2023, doi:10.32604/cmc.2023.030789.
110. Y. Jiang, P. Shen, L. Zeng, X. Zhu, D. Jiang, C. Chen, “Cancelable biometric schemes for euclidean metric and cosine metric”, Cyberse- curity, vol. 6, no. 1, p. 4, 2023, doi:10.1186/s42400-023-00137-0.
111. O. S. Faragallah, E. A. Naeem, W. El-Shafai, N. Ramadan, H. E.-d. H. Ahmed, M. M. A. Elnaby, I. Elashry, S. E. El-Khamy, F. E. A. El-Samie, “Efficient chaotic-baker-map-based cancelable face recognition”, Jour- nal of Ambient Intelligence and Humanized Computing, pp. 1–39, 2023, doi:10.1007/s12652-021-03398-0.
112. S. Yamamoto, H. Inaba, “Cancellable biometric authentication system by image style transfer”, “2023 IEEE 12th Global Confer- ence on Consumer Electronics (GCCE)”, pp. 794–797, IEEE, 2023, doi:10.1109/gcce59613.2023.10315291.
113. F. Turroni, D. Maltoni, R. Cappelli, D. Maio, “Improving fin- gerprint orientation extraction”, IEEE Transactions on Informa- tion Forensics and Security, vol. 6, no. 3, pp. 1002–1013, 2011, doi:10.1109/tifs.2011.2150216.
114. S. Li, A. C. Kot, “An improved scheme for full fingerprint reconstruc- tion”, IEEE Transactions on Information Forensics and Security, vol. 7, no. 6, pp. 1906–1912, 2012, doi:10.1109/tifs.2012.2212012.
115. J. Galbally, S. Marcel, J. Fierrez, “Image quality assessment for fake biometric detection: Application to iris, fingerprint, and face recogni- tion”, IEEE transactions on image processing, vol. 23, no. 2, pp. 710–724, 2013, doi:10.1109/tip.2013.2292332.
116. H.-W. Jung, J.-H. Lee, “Noisy and incomplete fingerprint classifi- cation using local ridge distribution models”, Pattern recognition, vol. 48, no. 2, pp. 473–484, 2015, doi:10.1016/j.patcog.2014.07.030.
117. S. Mathur, A. Vjay, J. Shah, S. Das, A. Malla, “Methodology for partial fingerprint enrollment and authentication on mobile devices”, “2016 International Conference on Biometrics (ICB)”, pp. 1–8, IEEE, 2016, doi:10.1109/icb.2016.7550093.
118. T. Chugh, S. S. Arora, A. K. Jain, N. G. Paulter, “Benchmarking fingerprint minutiae extractors”, “2017 International conference of the biometrics special interest group (BIOSIG)”, pp. 1–8, IEEE, 2017, doi:10.23919/biosig.2017.8053498.
119. G. S. E. Ekladious, R. Sabourin, E. Granger, “Learning global-local distance metrics for signature-based biometric cryptosystems”, Cryp- tography, vol. 1, no. 3, p. 22, 2017, doi:10.3390/cryptography1030022.
120. R. D. Labati, A. Genovese, E. Munoz, V. Piuri, F. Scotti, “A novel pore extraction method for heterogeneous fingerprint images using convolutional neural networks”, Pattern Recognition Letters, vol. 113, pp. 58–66, 2018, doi:10.1016/j.patrec.2017.04.001.
121. R. Gupta, M. Khari, D. Gupta, R. G. Crespo, “Fingerprint image enhancement and reconstruction using the orientation and phase reconstruction”, Information Sciences, vol. 530, pp. 201–218, 2020, doi:10.1016/j.ins.2020.01.031.
122. F. Liu, Y. Zhao, G. Liu, L. Shen, “Fingerprint pore matching us- ing deep features”, Pattern Recognition, vol. 102, p. 107208, 2020, doi:10.1016/j.patcog.2020.107208.
123. W. Yang, S. Wang, K. Yu, J. J. Kang, M. N. Johnstone, “Secure finger- print authentication with homomorphic encryption”, “2020 Digital Image Computing: Techniques and Applications (DICTA)”, pp. 1–6, IEEE, 2020, doi:10.1109/dicta51227.2020.9363426.
124. B. T. Ahmed, O. Y. Abdulhameed, “Fingerprint authentica- tion using shark smell optimization algorithm”, UHD Jour- nal of Science and Technology, vol. 4, no. 2, pp. 28–39, 2020, doi:10.21928/uhdjst.v4n2y2020.pp28-39.
125. T. Kim, Y. Oh, H. Kim, “Efficient privacy-preserving fingerprint- based authentication system using fully homomorphic encryption”, Security and Communication Networks, vol. 2020, pp. 1–11, 2020, doi:10.1155/2020/4195852.
126. M. Golec, S. S. Gill, R. Bahsoon, O. Rana, “Biosec: A biometric authen- tication framework for secure and private communication among edge devices in iot and industry 4.0”, IEEE Consumer Electronics Mag- azine, vol. 11, no. 2, pp. 51–56, 2020, doi:10.1109/mce.2020.3038040.
127. M. Lebcir, S. Awang, A. Benziane, “Reversible watermarking tech- nique for fingerprint authentication based on dct”, IOP Conference Series: Materials Science and Engineering, vol. 769, no. 1, p. 012070, 2020, doi:10.1088/1757-899x/769/1/012070.
128. D. Valdes-Ramirez, M. A. Medina-Pérez, R. Monroy, “An ensemble of fingerprint matching algorithms based on cylinder codes and mtriplets for latent fingerprint identification”, Pattern Analysis and Ap- plications, vol. 24, pp. 433–444, 2021, doi:10.1007/s10044-020-00911-7.
129. I. Joshi, R. Kothari, A. Utkarsh, V. K. Kurmi, A. Dantcheva, S. D. Roy, P. K. Kalra, “Explainable fingerprint roi segmentation using monte carlo dropout”, “Proceedings of the IEEE/CVF Winter Con- ference on Applications of Computer Vision”, pp. 60–69, 2021, doi:10.1109/wacvw52041.2021.00011.
130. P. Terhörst, A. Boller, N. Damer, F. Kirchbuchner, A. Kuijper, “Mide- con: Unsupervised and accurate fingerprint and minutia quality assessment based on minutia detection confidence”, “2021 IEEE International Joint Conference on Biometrics (IJCB)”, pp. 1–8, IEEE, 2021, doi:10.1109/ijcb52358.2021.9484404.
131. F. Pandey, P. Dash, D. Samanta, M. Sarma, “Asra: Automatic singular value decomposition-based robust fingerprint image alignment”, Multimedia Tools and Applications, vol. 80, pp. 15647–15675, 2021, doi:10.1007/s11042-021-10560-5.
132. A. Muhammed, N. C. Mhala, A. R. Pais, “A novel fingerprint template protection and fingerprint authentication scheme using visual se- cret sharing and super-resolution”, Multimedia Tools and Applications, vol. 80, no. 7, pp. 10255–10284, 2021, doi:10.1007/s11042-020-10095-1.
133. R. C. Contreras, L. G. Nonato, M. Boaventura, I. A. G. Boaventura, B. G. Coelho, M. S. Viana, “A new multi-filter framework with sta- tistical dense sift descriptor for spoofing detection in fingerprint authentication systems”, “International Conference on Artificial Intelligence and Soft Computing”, pp. 442–455, Springer, 2021, doi:10.1007/978-3-030-87897-9_39.
134. M. S. El_Tokhy, “Robust multimodal biometric authentication al- gorithms using fingerprint, iris and voice features fusion”, Journal of Intelligent & Fuzzy Systems, vol. 40, no. 1, pp. 647–672, 2021, doi:10.3233/jifs-200425.
135. A. Bedari, S. Wang, W. Yang, “A secure online fingerprint authenti- cation system for industrial iot devices over 5g networks”, Sensors, vol. 22, no. 19, p. 7609, 2022, doi:10.3390/s22197609.
136. A. Siswanto, A. Efendi, E. A. Kadir, “Fingerprint authentication in smart home environment based on embedded system”, “2022 International Conference on Electrical, Computer, Communications and Mechatronics Engineering (ICECCME)”, pp. 1–6, IEEE, 2022, doi:10.1109/iceccme55909.2022.9988711.
137. N. Pradeep, J. Ravi, “An revolutionary fingerprint authentication approach using gabor filters for feature extraction and deep learning classification using convolutional neural networks”, “Innovations in Electronics and Communication Engineering: Proceedings of the 9th ICIECE 2021”, pp. 349–360, Springer, 2022, doi:10.1007/978-981- 16-8512-5_38.
138. A. F. Y. Althabhawee, B. K. O. C. Alwawi, “Fingerprint recognition based on collected images using deep learning technology”, IAES International Journal of Artificial Intelligence, vol. 11, no. 1, p. 81, 2022, doi:10.11591/ijai.v11.i1.pp81-88.
139. P. Singh, H. Samuel, F. Jaafar, D. Ameyed, “Enhancing biometric security with combinatorial and permutational multi-fingerprint authentication strategies”, “2022 IEEE Intl Conf on Dependable, Autonomic and Secure Computing, Intl Conf on Pervasive Intel- ligence and Computing, Intl Conf on Cloud and Big Data Com- puting, Intl Conf on Cyber Science and Technology Congress (DASC/PiCom/CBDCom/CyberSciTech)”, pp. 1–7, IEEE, 2022, doi:10.1109/picom/cbdcom/cy55231.2022.9927942.
140. Y. Chen, Y. Yu, L. Zhai, “ InfinityGauntlet : Expose smartphone fingerprint authentication to brute-force attack”, “32nd USENIX Security Symposium (USENIX Security 23)”, pp. 2027–2041, 2023.
141. A. Popli, S. Tandon, J. J. Engelsma, A. Namboodiri, “A unified model for fingerprint authentication and presentation attack detection”, “Handbook of Biometric Anti-Spoofing: Presentation Attack De- tection and Vulnerability Assessment”, pp. 77–99, Springer, 2023, doi:10.1109/ijcb52358.2021.9484382.
142. H. Choi, S. Woo, H. Kim, “Blind-touch: Homomorphic encryption- based distributed neural network inference for privacy-preserving fingerprint authentication”, arXiv preprint arXiv:2312.11575, 2023, doi:10.1609/aaai.v38i20.30200.
143. E. Marasco, M. Albanese, V. V. R. Patibandla, A. Vurity, S. S. Sri- ram, “Biometric multi-factor authentication: On the usability of the fingerpin scheme”, Security and Privacy, vol. 6, no. 1, p. e261, 2023, doi:10.1002/spy2.261.
144. R. Deshmukh, P. Yannawar, “Avao enabled deep learning based person authentication using fingerprint”, “First International Con- ference on Advances in Computer Vision and Artificial Intelligence Technologies (ACVAIT 2022)”, pp. 327–346, Atlantis Press, 2023, doi:10.2991/978-94-6463-196-8_26.
145. D. Peralta, I. Triguero, R. Sanchez-Reillo, F. Herrera, J. Benitez, “Fast fingerprint identification for large databases”, Pattern Recognition, vol. 47, no. 2, p. 588–602, 2014, doi:10.1016/j.patcog.2013.08.002.
146. M. Lastra, J. Carabaño, P. D. Gutiérrez, J. M. Benítez, F. Herrera, “Fast fingerprint identification using gpus”, Information Sciences, vol. 301, pp. 195–214, 2015, doi:10.1016/j.ins.2014.12.052.
147. Y. Su, J. Feng, J. Zhou, “Fingerprint indexing with pose constraint”, Pattern Recognition, vol. 54, pp. 1–13, 2016, doi:10.1016/j.patcog.2016.01.006.
148. J. Khodadoust, A. M. Khodadoust, “Fingerprint indexing based on minutiae pairs and convex core point”, Pattern Recognition, vol. 67, pp. 110–126, 2017, doi:10.1016/j.patcog.2017.01.022.
149. D. Song, Y. Tang, J. Feng, “Aggregating minutia-centred deep con- volutional features for fingerprint indexing”, Pattern Recognition, vol. 88, pp. 397–408, 2019, doi:10.1016/j.patcog.2018.11.018.
150. J. J. Engelsma, K. Cao, A. K. Jain, “Learning a fixed-length fingerprint representation”, IEEE transactions on pattern analy- sis and machine intelligence, vol. 43, no. 6, pp. 1981–1997, 2019, doi:10.1109/tpami.2019.2961349.
151. U. U. Deshpande, V. Malemath, S. M. Patil, S. V. Chaugule, “Cnnai: a convolution neural network-based latent fingerprint matching using the combination of nearest neighbor arrange- ment indexing”, Frontiers in Robotics and AI, vol. 7, p. 113, 2020, doi:10.3389/frobt.2020.00113.
152. V. Anand, V. Kanhangad, “Pore-based indexing for fingerprints acquired using high-resolution sensors”, Pattern Analysis and Appli- cations, vol. 23, pp. 429–441, 2020, doi:10.1007/s10044-019-00805-3.
153. H. P. Singh, P. Dimri, S. Tiwari, M. Saraswat, “Segmentation tech- niques through machine based learning for latent fingerprint index- ing and identification”, 2020, doi:10.56042/jsir.v79i3.68640.
154. I. Pérez-Sánchez, B. Cervantes, M. A. Medina-Pérez, R. Monroy, O. Loyola-González, S. García, F. Herrera, “An indexing algorithm based on clustering of minutia cylinder codes for fast latent fin- gerprint identification”, IEEE Access, vol. 9, pp. 85488–85499, 2021, doi:10.1109/access.2021.3088314.
155. Y. Xu, Y. Lu, F. Chen, G. Lu, D. Zhang, “High resolution fingerprint retrieval based on pore indexing and graph comparison”, IEEE Transactions on Information Forensics and Security, vol. 17, pp. 226–236, 2021, doi:10.1109/tifs.2021.3139219.
156. J. M. S. Soares, L. Barbosa, P. A. L. Rego, R. P. Magalhães,
     J. A. F. de Macêdo, “Using inverted index for fingerprint search”, Journal of Information and Data Management, vol. 12, no. 5, 2021, doi:10.5753/jidm.2021.1918.
157. G. Arora, S. Kalra, A. Bhatia, K. Tiwari, “Palmhashnet: Palm- print hashing network for indexing large databases to boost identification”, IEEE Access, vol. 9, pp. 145912–145928, 2021, doi:10.1109/access.2021.3123291.
158. R. Balasundaram, G. F. Sudha, “Retrieval performance analysis of multibiometric database using optimized multidimensional spectral hashing based indexing”, Journal of King Saud University- Computer and Information Sciences, vol. 33, no. 1, pp. 110–117, 2021, doi:10.1016/j.jksuci.2018.02.003.
159. D. Osorio-Roig, T. Schlett, C. Rathgeb, J. Tapia, C. Busch, “Exploring quality scores for workload reduction in biometric identification”, “2022 International Workshop on Biometrics and Forensics (IWBF)”, pp. 1–6, IEEE, 2022, doi:10.1109/iwbf55382.2022.9794533.
160. G. Arora, A. Singh, A. Nigam, H. M. Pandey, K. Tiwari, “Fkpindexnet: An efficient learning framework for finger-knuckle-print database indexing to boost identification”, Knowledge-Based Systems, vol. 239, p. 108028, 2022, doi:10.1016/j.knosys.2021.108028.
161. P. Drozdowski, F. Stockhardt, C. Rathgeb, C. Busch, “Signal-level fusion for indexing and retrieval of facial biometric data”, IET Biometrics, vol. 11, no. 2, pp. 141–156, 2022, doi:10.1049/bme2.12063.
162. D. Osorio-Roig, L. J. Gonzalez-Soler, C. Rathgeb, C. Busch, “Privacy-preserving multi-biometric indexing based on fre- quent binary patterns”, arXiv preprint arXiv:2310.03091, 2023, doi:10.1109/tifs.2024.3386310/mm1.
163. S. A. Grosz, A. K. Jain, “Afr-net: Attention-driven fingerprint recog- nition network”, IEEE Transactions on Biometrics, Behavior, and Identity Science, 2023, doi:10.1109/tbiom.2023.3317303.
164. M. Kumar, D. Kumar, “An efficient gravitational search decision forest approach for fingerprint recognition”, Kuwait Journal of Science, vol. 50, no. 2A, 2023, doi:10.48129/kjs.20635.
165. A. K. Jain, S. S. Arora, L. Best-Rowden, K. Cao, P. S. Sudhish, A. Bhat- nagar, Y. Koda, “Giving infants an identity: Fingerprint sensing and recognition”, “Proceedings of the Eighth International Conference on Information and Communication Technologies and Development”, pp. 1–4, 2016, doi:10.1145/2909609.2909612.
166. A. K. Jain, S. S. Arora, K. Cao, L. Best-Rowden, A. Bhatnagar, “Fingerprint recognition of young children”, IEEE Transactions on Information Forensics and Security, vol. 12, no. 7, pp. 1501–1514, 2016, doi:10.1109/tifs.2016.2639346.
167. J. J. Engelsma, D. Deb, K. Cao, A. Bhatnagar, P. S. Sudhish, A. K. Jain, “Infant-id: Fingerprints for global good”, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 44, no. 7, pp. 3543–3559, 2021, doi:10.1109/tpami.2021.3057634.
168. I. Widiatmika, I. N. Piarsa, A. Syafiandini, “Recognition of the baby footprint characteristics using wavelet method and k-nearest neigh- bor (k-nn)”, Lontar Komputer: Jurnal Ilmiah Teknologi Informasi, vol. 12, no. 1, pp. 41–52, 2021, doi:10.24843/lkjiti.2021.v12.i01.p05.
169. H. I. Yoshinori Koda, “Fundamental study of neonate fin- gerprint recognition using fingerprint classification”, 2022, doi:10.1109/biosig55365.2022.9897017.
170. N. Nelufule, Y. Moolla, S. Ntshangase, A. de Kock, “Biometric recog- nition of infants using fingerprints”, Tech. rep., EasyChair, 2023, doi:10.1109/ictas56421.2023.10082749.
171. T. O. Odu, T. Ogunfunmi, M. O. Olaniyan, I. A. Samuel, “Multi- instance contingent fusion for the verification of infant fingerprints”, “World Conference on Information Systems for Business Manage- ment”, pp. 197–207, Springer, 2023, doi:10.1155/2024/7728707.
172. M. Bahzad, L. M. Labib, M. Elhosseini, M. Badawy, “M2brtpc: A novel modified multimodal biometric recognition for toddlers and pre-school children approach”, Mansoura Engineering Journal, vol. 48, no. 5, p. 5, 2023, doi:10.58491/2735-4202.3069.
173. M. Shabil, H. Fadewar, “Fingerprint recognition of newborns and toddlers using pre-trained model under convolution neural net- works”, Journal of Data Acquisition and Processing, vol. 38, no. 2, p. 234, 2023, doi:10.5281/zenodo.7766329.
174. K. Rajaram, N. B. Amma, S. Selvakumar, “Convolutional neural network based children recognition system using contactless finger- prints”, International Journal of Information Technology, vol. 15, no. 5, pp. 2695–2705, 2023, doi:10.1007/s41870-023-01306-7.
175. M. Kazi, K. Kale, R. S. Mehsen, A. Mane, V. Humbe, Y. Rode, S. Dab- hade, N. Bansod, A. Razvi, P. Deshmukh, “Face, fingerprint, and signature based multimodal biometric system using score level and decision level fusion approaches”, IETE Journal of Research, pp. 1–20, 2023, doi:10.1080/03772063.2023.2217784.
176. S. A. El_Rahman, A. S. Alluhaidan, “Enhanced multimodal biometric recognition systems based on deep learning and traditional methods in smart environments”, Plos one, vol. 19, no. 2, p. e0291084, 2024, doi:10.1371/journal.pone.0291084.
177. H. Byeon, V. Raina, M. Sandhu, M. Shabaz, I. Keshta, M. Soni, K. Ma- trouk, P. P. Singh, T. Lakshmi, “Artificial intelligence-enabled deep learning model for multimodal biometric fusion”, Multimedia Tools and Applications, pp. 1–24, 2024, doi:10.1007/s11042-024-18509-0.
178. J. Priyani, P. Nanglia, P. Singh, V. Shokeen, A. Sharma, “Hgssa-bi lstm: A secure multimodal biometric sensing using optimized bi- directional long short-term memory with self-attention”, ECS Sensors Plus, vol. 3, no. 1, p. 011401, 2024, doi:10.1149/2754-2726/ad1b3a.
179. J. Samatha, G. Madhavi, “Securesense: Enhancing person verification through multimodal biometrics for robust authentication”, Scalable Computing: Practice and Experience, vol. 25, no. 2, pp. 1040–1054, 2024, doi:10.12694/scpe.v25i2.2524.
180. D. A. Reid, M. S. Nixon, S. V. Stevenage, “Soft biometrics: Human identification using comparative descriptions”, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 36, no. 6, pp. 1216–1228, 2014, doi:10.1109/tpami.2013.219.

Citations by Dimensions

Citations by PlumX

Crossref Citations

This paper is currently not cited.

No. of Downloads per Month

No. of Downloads per Country

Tab Content