FIGURE 2 - available via license: Creative Commons Attribution 4.0 International
Content may be subject to copyright.
Source publication
Fingerprint-based recognition is widely deployed in different domains. However, the traditional fingerprint recognition systems are vulnerable to presentation attack, which utilizes an artificial replica of the fingerprint to deceive the sensors. In such scenarios, Fingerprint Liveness Detection (FLD) is required to ensure the actual presence of a...
Context in source publication
Context 1
... purpose of fingerprint matching is to determine whether the two fingerprint images come from the same finger by calculating the similarity of two images. Minutiae based fingerprint matching algorithms are currently widely employed, and the specific minutiae types are limited to two: endings and bifurcations (Figure 2). They can be described using parameters such as coordinates, direction and type. ...
Similar publications
Fingerprint liveness detection is an essential module for an accurate and reliable fingerprint identification system. In this paper, a Densely Connected Convolutional Network (DenseNet) is used for fingerprint liveness detection and the genetic algorithm is adopted to optimize the DenseNet structure. Firstly, all images in the experimental database...
Citations
... However, they need extensive training and involve high computational complexity [51]. Zhang et al. [68,69] calculated similarity score for fingerprint liveness detection using Octantal Nearest-Neighborhood Structure (ONNS) and the score was obtained using Slim-Residual Network. Logistic Regression classifiers give an accuracy of 96.88%. ...
... 46 (3) LivDet-2015 Digital Persona 86.52 (7) 97.59 (1) 96.59 (2) 83.42 (6) 84.67 (5) 85.29 (4) 89.20 (3) LivDet-2015 CrossMatch 86.89 (7) 97. 68 (1) 96.28 (2) 84.29 (6) 85.54 (5) 86. 30 Content courtesy of Springer Nature, terms of use apply. ...
Presentation attacks that make the biometric systems vulnerable has become a growing concern in recent years keeping in view its widespread applications in the field of banking, medical, security systems etc. For instance, textured contact lenses, high-quality printouts and fabricated synthetic materials spoof the iris texture and fingerprints that lead to increase in false rejection. Till now, extensive work has been done on global features. However, this paper proposed local features with invariance properties. Thus, the paper proposes detection of spoofing attacks in which local features are extracted for micro-textural analysis with properties of invariance to scale, rotation and translation. The features are encoded using Lehmer code and transformed into histograms that act as feature descriptors for classification. The top 4 features are selected using Friedman test. Experiments are simulated on iris spoofing databases: IIITD-Contact Lens, IIITD-Iris Spoofing, Clarkson-2015, Warsaw-2015and fingerprint spoofing databases: LivDet-2013 and LivDet-2015. Results have been validated through intra-sensor, inter-sensor, cross-sensor and cross-material. In case of IIITD-CLI, an EER of 1.36% and an ACER of 1.45% is obtained. For IIS, 0.94% of EER and 1.61% of ACER is observed. For Clarkson database, 0.79% of EER and 2.10% of ACER is obtained. An ACER of 0.57% is obtained for LivDet-2013 and 0.47% for LivDet-2015.
... Their evaluation results show that the K-nearest neighbours algorithm outperformed the other algorithms in terms of accuracy, while the Euclidean distance algorithm was found to be the least complex computationally. Another method that aims to increase the accuracy and robustness of fingerprint recognition is proposed in Zhang et al. (2020). This method combines the scores obtained from a fingerprint matcher and a liveness detector. ...
... Bio-WISE allowed also exploring the performance as function of the attack probability. However, the current version of Bio-WISE and, in general, the findings of Ref. [12] and related previous works [6]- [10], [13], [14] do not allow assessing for which PAD's operational points the overall GAR degradation can be still acceptable, with the advantage of handling presentation attacks. The key question is: what is the best way to embed a PAD into a recognition system so that the final product is robust to spoofing (IAPAR low enough), without suffering from significantly reduced gen-uine recognition accuracy (GAR still acceptable)? ...
... A noteworthy contribution to the field has been made by the last editions of LivDet, aimed at promoting the development of integrated systems by providing a common platform for researchers to evaluate and compare the performance of their algorithms. The solutions presented employ score-level fusion to generate a unified metric score [14]. In general, all these works follow a similar approach, utilizing two independent architectures to carry out the presentation attack detection and recognition task and implementing a fusion method at the output level. ...
The assessment of the fingerprint PADs embedded into a comparison system represents an emerging topic in biometric recognition. Providing models and methods for this aim helps scientists, technologists, and companies to simulate multiple scenarios and have a realistic view of the process’s consequences on the recognition system. The most recent models aimed at deriving the overall system performance, especially in the sequential assessment of the fingerprint liveness and comparison pointed out a significant decrease in Genuine Acceptance Rate (GAR). In particular, our previous studies showed that PAD contributes predominantly to this drop, regardless of the comparison system used. This paper’s goal is to establish a systematic approach for the “trade-off” computation between the gain in Impostor Attack Presentation Accept Rate (IAPAR) and the loss in GAR mentioned above. We propose a formal “trade-off” definition to measure the balance between tackling presentation attacks and the performance drop on genuine users. Experimental simulations and theoretical expectations confirm that an appropriate “trade-off” definition allows a complete view of the sequential embedding potentials.
... The Unimodal Biometric System (UBS) is a biometric authentication system that depends on a single source of biometric data, such as a single fingerprint or face scan. Although there have been significant improvements recently, there are still many drawbacks to accurate UBS authentication, including noise in collected data, intra-class differences, inter-class correlation, non-universality, and unauthorized attacks [1,2]. The abilities of UBS are limited and inadequate for comprehensive evidence verification of a person. ...
... Zhang et al. [1] presented a score-level integration method that combined the liveness detection and fingerprint matching scores to generate the last integrated score for figuring out if the fingerprint to be authenticated is truly a living fingerprint. The simulation results reveal that despite the fictitious fingerprints in the test data set being created using biometric data that was unique from that used in the training set, the integrated model proffered achieved a remarkable efficiency with accuracy of 96.88%. ...
... The RA increased by 17.31%, FPR reduced by 20.21%, FNR reduced by 18.02%, respectively for GSA-based MFAS with respect to [9] values. Also, there was 0.43% increase of accuracy for GSA-based MFAS with respect to [1] evaluation. ...
Multi-instance fusion of fingerprint authentication system at score level overcomes a few of the shortcomings of a Unimodal Biometric System (UBS) and enhanced the efficiency of the system. However, due to loss of information at higher levels, the features fused at the score level are confined in comparison to feature level fusion and could lead to poor performance. In this study, multi-instance fusion of fingerprints was done at feature level using Gravitational Search Algorithm (GSA) to select and combine minimal relevant informative texture features subsets from multi-instances of fingerprint and considerably improves the performance of the system. The approach was validated by creation of multi-instances of fingerprint database acquired locally from 150 subjects in an uncontrolled environment and texture based feature extraction was considered and classification of fused texture feature was done using back propagation neural network. The results show that the presented technique was effective in subject authentication with accuracy of 97.09%, indicating that it can successfully secure fingerprint authentication systems from unauthorized attacks.
... The aforementioned study by Kumar and Om was also discussed in terms of its robustness against adversarial attacks, and vulnerabilities were uncovered by the article of Inam ul haq et al. (2021), where they proposed a novel scheme that is resilient to key compromise impersonation. Besides, specific spoof and attack cases towards fingerprint recognition schemes were discussed by Uludag and Jain (2004), and Zhang et al. (2020). In a different input domain, Ming et al. (2020) outlined adversary activities targeting face recognition methods and reported the prospective antispoofing methods. ...
... Many studies showed that feature fusion can improve the accuracy of object detection (Dong et al., 2022) or recognition (Eleyan, 2023). In score fusion, the match scores are combined using certain rules such as the sum rule after applying normalization of each classifier's scores (Aizi and Ouslim, 2022;Zhang et al., 2020). Finally, decision fusion involves combining the outputs of multiple decision-making algorithms using different rules, such as majority voting, to obtain a reliable decision. ...
... In the study [25], fingerprint matching score and fingerprint liveness detection score are combined by score level fusion mechanism to avoid spoofing attacks. This approach is one kind of advancement in the traditional way of fingerprint recognition. ...
... The aforementioned study by Kumar et al. was also discussed in terms of its robustness against adversarial attacks, and vulnerabilities were uncovered by the article of Inam ul haq et al. (2021), in which they proposed a novel scheme that is resilient to key compromise impersonation. Besides, specific spoof and attack cases towards fingerprint recognition schemes were discussed by Uludag et al. (2004), andZhang et al. (2020). In a different input domain, Ming et al. (2020) outlined adversary activities targeting face recognition methods and reported the prospective anti-spoofing methods. ...
The final version is published by Elsevier, in "Expert Systems with Applications" and available from https://doi.org/10.1016/j.eswa.2023.121323
... Various other approaches have been proposed to detect liveness in a fingerprint, including using a score-level fusion [20], reaching 96.88% accuracy, using automatic template updating using the fusion of ECG and Fingerprint [21], reaching 97.4% accuracy, using a two-layer parallel SVM network based on aggregated local descriptors [22], reaching 95.32% accuracy, and using SVM [23], reaching 100% accuracy. Table 1 summarizes the methods described in the literature and the accuracy reached in differentiating between fake and real fingerprints. ...
Fingerprints have been used for decades to verify the identity of an individual for various security reasons. Attackers have developed many approaches to deceive a fingerprint verification system, ranging from the sensor level, where gummy fingers are created, to gaining access to the decision-maker level, where the decision is made based on low matching criteria. Even though fingerprint sensor-level countermeasures have developed advanced metrics to detect any attempt to dupe the system, attackers still manage to outwit a fingerprint verification system. In this paper, we present the Micro-behavioral Fingerprint Analysis System (MFAS), a system that records the micro-behavior of the user’s fingertips over time as they are placing their fingerprint on the sensor. The system captures the stream of ridges as they are formed while placed on a sensor to combat the attacks that deceive the sensor. An experiment on 24 people was conducted, wherein the fingerprints and the behavior of the fingertip as it is placed were collected. Subsequently, a gummy finger was created to try to fool the system. Further, a legitimate user was chosen to participate in an experiment that mimicked an attempt to use their fingertip unwillingly to detect coerced fingerprint placement. After applying the micro-behavior, the system reported 100% true positives and 0% false-negatives when providing legitimate vs. gummy-based fingerprints to authenticate a malicious user. The system also reported a 100% accuracy in differentiating between a voluntary and a coerced fingerprint placement. The results improve the fingerprint robustness against attacks on a fingerprint sensor by factoring in micro-behavior, thus helping to overcome fake and coerced fingerprint attacks.
... In this case, the police extracted an incomplete fingerprint at the scene. Based on this fingerprint, the American police mistakenly identified others as the perpetrator [5]. In 2014, the Miami Police Department of the United States conducted statistics on fingerprint error identification, and the results showed that the false true rate of fingerprint identification was 3.0%, and the false error rate was 7.5%, which shows that the qualitative fingerprint identification conclusion is not completely reliable [6]. ...
... After determining the depth of the Backbone network, on the basis of YOLOv5_16, Backbone adds the SPP pyramid pooling module, and Neck uses the FPN and PAN structure to fuse the features of the 8-fold and 16-fold downsampling layers, naming the model YOLOv5s_B. In order to explore the optimal pooling effect of SPP, four common SPP pooling kernels are tested in this paper: (3,5,7), (5,7,9), (7,9,13) and (9,11,13), named YOLOv5s_B_a, YOLOv5s_B_b, YOLOv5s_B_c, and YOLOv5s_B_d, respectively. Its structure is shown in Figure 9. ...
... After determining the depth of the Backbone network, on the basis of YOLOv5_16, Backbone adds the SPP pyramid pooling module, and Neck uses the FPN and PAN structure to fuse the features of the 8-fold and 16-fold downsampling layers, naming the model YOLOv5s_B. In order to explore the optimal pooling effect of SPP, four common SPP pooling kernels are tested in this paper: (3,5,7), (5,7,9), (7,9,13) and (9,11,13), named YOLOv5s_B_a, YOLOv5s_B_b, YOLOv5s_B_c, and YOLOv5s_B_d, respectively. Its structure is shown in Figure 9. ...
The quantitative identification technology based on the statistical law of fingerprint features has become a new research difficulty and focus, and the automatic detection and classification of fingerprint features are the basis for realizing automatic fingerprint feature statistics. In this paper, a YOLO-based fingerprint feature detection method was proposed. First, a fingerprint feature dataset was established, which contained a total of 4,000 annotated fingerprint images; then, according to the characteristics of small size and dense distribution of fingerprint feature points, the YOLO network structure was improved, the original large target feature detection layer by 32-fold downsampling was deleted, and a new small feature fusion layer was added; the FPN, PAN, and SPP structures were used to achieve local and global feature extraction through multiple-scale fusion methods; finally, the SE channel attention mechanism module was added to effectively enhance the model robustness and detection ability of dense small objects. The experimental results show that compared with the improved FP-YOLO model of the original model, when the detection speed is basically unchanged, the mAP0.5 value is increased from 93.0% to 97.4%, and the weight is reduced by 3/4.
