VeriFace: Face Recognition Engine

Face liveness detection systems are commonly trained on real-world facial images, where genuine faces and spoofattempts often appear quite similar. However, there has been limited research on identifying live faces using a blend of realfacial photos and synthetic images generated by deep convolutional neural networks (CNNs). As facial recognition becomesmore widely used for biometric authentication, it's crucial that these systems not only recognize genuine users but also resistspoofing attempts—such as using printed images, digital displays, or replayed videos.To enhance spoof prevention, this approach introduces a liveness detection method that integrates a CNN-based classifierwith an activity-based analysis module. Specifically, it monitors facial behaviors like eye blinking and lip movement to detectsigns of life. These features are essential because conventional spoofing methods struggle to replicate such subtle, spontaneousactions.The solution is built using Python and combines two modules: one powered by CNN for feature classification, and anotherthat tracks eye and mouth movements. Together, they form a robust defense against various spoofing attacks, including thoseinvolving posters, silicone masks, or smartphone screens.Moreover, the study proposes an adaptive fusion layer that harmonizes convolutional features from both real faces andCNN-generated synthetic faces during the training process. This fusion enables the model to generalize better across differentscenarios.Extensive evaluations using prominent face anti-spoofing datasets—such as CASIA, OULU, and Replay-Attack—demonstrate that the proposed method achieves superior performance in both intra-database and cross-database testingenvironments, outperforming several existing state-of-the-art approaches.