Anti-Spoofing: Liveness Detection System

Abstract

Face recognition systems became more susceptible to presentation attacks by digital screens, printed images, and 3D masks [3]. This paper introduces a full-fledged anti-spoofing solution based on the YOLO (You Only Look Once) frameworktoidentifyandthwart suchattemptsatspoofinginreal-time[14]. Oursystemintegrates effective object detection features with custom liveness evaluation features to form an effective security layer for biometric authentication systems. Experimental results show high accuracy in distinguishing between real users and spoofing attempts with real-time performance appropriate for practical use [4]. The study points out the efficiency of feature extraction from biometric informationusingCNNs [16] and the capacity ofTransformers to model global dependencies for improvedspoof detection [11]. By combining these approaches, the study seeks to enhance the accuracy and reliability of liveness detection, mitigating vulnerabilities in biometric authentication systems [9].

Introduction

The study addresses biometric authentication's vulnerability to spoofing attacks such as masks, photos, and synthetic fingerprints. It emphasizes the importance of liveness detection to differentiate genuine users from spoof attempts. The research compares Convolutional Neural Networks (CNNs) enhanced with self-supervised learning to Vision Transformers (ViTs), finding CNNs achieve higher accuracy and recall.

The literature review highlights the evolution of biometric anti-spoofing methods from simple techniques to advanced deep learning models like CNNs and Transformers. CNNs excel at extracting spatial features, while Transformers leverage self-attention to capture global image context, improving detection of sophisticated spoofing attacks such as 3D masks. Hybrid models combining CNNs with machine learning classifiers show promise.

The authors propose a novel liveness detection system integrating YOLO for fast, real-time face detection with feature extraction modules analyzing texture, color, frequency, and motion cues to detect spoofing. The system is deployed as a web app using Streamlit and WebRTC for accessible, hardware-independent operation.

Experiments on diverse datasets demonstrate high accuracy (96.8%), low false acceptance/rejection rates, and real-time processing (41 FPS). The system performs well across attack types, lighting conditions, distances, and demographics, outperforming existing methods.

Limitations include challenges detecting advanced 3D masks, low-light performance degradation, moderate computational demands, and potential vulnerability to adversarial attacks. Ethical concerns such as privacy, bias, transparency, and system security are also discussed.

Future work includes improving 3D mask detection using specialized sensors, exploring multi-modal biometric approaches, and enhancing robustness against emerging spoofing methods.

Conclusion

This research presented a novel liveness detection system for facialanti-spoofingbasedontheYOLOobjectdetection framework. Our approach combines YOLO\'s efficient real- time detection capabilities with specialized features for liveness assessment, creating a robust solution for distinguishing between genuine faces and presentation attacks. The implementation asa web application using Streamlit and WebRTC features makes the system accessible through standard browsers, facilitating practical deployment [11][14]. Theexperimentalresultsdemonstratethatoursystemachieves highaccuracy (96.8%)with lowerror rates (HTERof 3.15%) while maintaining real-time performance (41 FPS).This makesitsuitable forpracticaldeploymentinsecurity-critical applicationsthatrequirereliablebiometricauthentication[11][14].

References

[1] Redmon,J.,&Farhadi,A.(2018).YOLOv3:AnIncrementalImprovement.arXivpreprint arXiv:1804.02767. [2] Liu, L., Ouyang, W., Wang, X., Fieguth, P., Chen, J., Liu, X., &Pietikäinen,M. (2020). Deep Learning for Generic Object Detection:A Survey. InternationalJournal of Computer Vision, 128(2), 261-318. [3] Zhang,S.,Yan,Y.,Luo,M.,Lei,Z.,Li,S.Z.,&Jain,A.K.(2020).A [4] SurveyonFaceAnti-Spoofing:Algorithms,Performance,andFutureTrends.arXivpreprintarXiv:2007.07721. [5] Patel,K.,Han,H.,&Jain,A.K.(2016).Cross-databasefaceantispoofingwithrobustfeaturerepresentation.InChineseConferenceonBiometricRecognition (pp. 611-619). Springer. [6] Jha,S.,&Sharma,S.(2021).ComparativeAnalysisofCNNandTransformer-basedArchitecturesforBiometricAuthentication.IEEEAccess,9,10874-10890. [7] George, A.,&Marcel,S.(2019).Deeppixel-wisebinarysupervisionforfacepresentationattackdetection.InternationalConferenceonBiometricSpecialInterest Group (BIOSIG), 1-6. [8] Agarwal,A.,&Bansal,R.(2016).Qualitymetrics-basedfingerprintlivenessdetection.PatternRecognitionLetters,84,175-185. [9] Frassetto,T.,Ortiz,M.,&Varela,C.(2015).Random-weightCNNsforfingerprintanti-spoofingwithsupportvectormachines.IEEETransactionsonInformationForensics and Security, 10(9), 1840-1852. [10] Sun, Y., Zheng, W., Liu, C., & Yin, Y. (2021). 3D Mask AttackDetection Using Convolutional Neural Networks and DepthEstimation. Pattern Recognition,121,108162. [11] Dong,J.,Yao,Y.,Yang,X.,&Liu,T.(2020).Faceanti-spoofingwithmulti-channelfeaturefusionandattentionmechanism.IEEETransactionsonBiometrics, Behavior, and Identity Science, 2(4), 401-412. [12] Li, H.,He, H., & Tan, Z. (2021). A Transformer-based approach forreal-time biometric authentication. IEEE Transactions on NeuralNetworks and LearningSystems, 32(3), 1560-1572. [13] Wang,Z.,&Liu,W.(2018).Facerecognitionunderspoofingattacks:Asurveyandnewinsights.ACMComputingSurveys(CSUR),51(4),1-36. [14] Arashloo,S.R.,Kittler,J.,&Christmas,W.(2017).Ananomalydetectionapproachtofacespoofingdetection:Anewformulationandevaluationprotocol. IEEE Transactions on Information Forensicsand Security, 12(10), 2436-2450. [15] Liu,Y.,Sun,L.,&Wei,X.(2022).YOLO-BasedFaceDetectionandFeatureExtractionforReal-TimeAuthenticationSystems.IEEEAccess,10,58632-58645. [16] Chingovska,I.,Anjos,A.,&Marcel,S.(2012).Ontheeffectivenessoflocalbinarypatternsinfaceanti-spoofing.InInternationalConferenceonBiometrics (ICB), 1-6. [17] Menotti,D.,Chiachia,G.,Pinto,A.,Schwartz,W.R.,Pedrini,H.,&Rocha,A.(2015).Deeprepresentationsforiris,face,andfingerprintspoofingdetection. IEEE Transactions on Information Forensics andSecurity, 10(4), 864-879. [18] Bao,W.,Li,H.,Li,N.,&Jiang,W.(2009).Alivenessdetectionmethodforface recognition based onoptical flowfield.In IEEE InternationalConference onImage Analysis and Signal Processing (pp. 233-236). [19] Zhang,Z., Yan, S., & Lei, Z. (2019).A comprehensive study on faceanti-spoofing with an ensemble of deep learning techniques. PatternRecognition,85,285-296. [20] Kumar, A., &Bhavsar,A.(2021). Ahybrid deep learning-basedapproachfor presentationattack detectionin facerecognitionsystems. JournalofVisual Communication and Image Representation,78, 103196. [21] Gragnaniello, D.,Sansone,C.,& Verdoliva, L. (2015).Fingerprintliveness detection based on Weberlocalimage descriptor.IEEESignal ProcessingLetters, 22(2), 176-180

Copyright

Copyright © 2025 Sharad Gupta, Yash Swaroop, Pankaj Kumar, Anuj Singh. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.