Deep Learning for Biometric Attack Detection and Recognition

Integration of biometric systems is essential for developing effective security and authentication strategies. These systems are nevertheless vulnerable to presentation attacks, in which nefarious actors compromise the integrity of the system by introducing fake biometric features. By enabling accurate presentation attack detection (PAD) and enhancing biometric recognition, deep learning has the potential to be a promising solution to this problem. This work contributes to the improvement of security measures by providing a thorough overview of recent developments in biometric PAD and recognition through the use of deep learning approaches. Video attacks stand out among the different methods used to trick face recognition systems as one of the most common, useful, and simple techniques. This study explores the area of face liveness detection in video attacks with a particular emphasis on determining the veracity of biometric information. The collection includes spoofing attack instances where frames were taken from movies to mimic the spoofing effort as well as instances taken from live facial footage. To categorise these cases, the research makes use of the Resnet-50 deep learning technique. To reach a final verdict, a majority voting system is used to combine several decisions. The paper also evaluates the spoofing attempts made using the movies from the Replay-Attack dataset, providing insight into the effectiveness of the suggested method.