CNN Face Live Detection Algorithm Based on Binocular Camera

Abstract

Abstract In this paper, a convolutional neural network (CNN) detection analysis is performed for live face detection by binocular cameras, and a binocular stereo matching network with fused edge detection is designed and implemented to target the quality of image details and parallax prediction at edges. Experiments show that the random sample pair confusion loss function can effectively improve the accuracy and generalization of the face live detection algorithm; the multi-task training approach can improve the performance of both faces live detection and face recognition; the algorithm shows excellent performance in both faces live detection and face recognition, especially the generalization of face live detection is greatly improved. A pre-trained convolutional neural network is used to extract features, and a content loss function and a domain loss function are designed to measure the feature distance between two images, and a feedforward neural network is trained as an image transformation network to migrate samples to the same domain. Experiments show that the algorithm can reduce the feature differences between the face live detection data of the two domains and can be used to improve the generalization of the face live detection algorithm.