Single-Shot Face Anti-Spoofing for Dual Pixel Camera

In this study, we propose a neural network-based face anti-spoofing algorithm using dual pixel (DP) sensor images. The proposed algorithm has two stages: depth reconstruction and depth classification. The first network takes a DP image pair as input and generates a depth map with a baseline of approximately 1 mm. Then, the classification network is trained to distinguish real individuals and planar attack shapes to produce a binary output. A DP image is utilized to estimate the depth map; thus, the proposed face anti-spoofing method is simple and robust. Experimental results demonstrate that the generated depth map helps distinguish real human faces from nonface attack, including images recaptured from photos or screens. The proposed algorithm achieves better anti-spoofing performance compared with other stereo and phase-based depth estimation schemes.

Face anti-spoofing is a crucial task for the security of face recognition, face detection, and verification systems. Previous approaches built models based on images or image sequences. However, with the rapid development of smartphones and cameras, dual pixel (DP) sensors are being utilized in several smartphone models [1]. In addition to fast and sharp autofocus, better image reproduction, and better object detection [2], DP technology provides useful information for various photo-related applications. For example, a DP image pair is captured separately and considered to be images with an extremely short baseline (less than 1 mm). With these images, subpixel disparity maps can be predicted, and the depth information of images can be recovered [3]. In this study, we demonstrate that depth information from DP cameras can be effectively used in face anti-spoofing. In particular, we propose an end-to-end neural network architecture to predict the depth map for DP cameras, which is applied to face anti-spoofing.