A Low-cost Life Sign Detection Method based on Time Series Analysis of Facial Feature Points

摘要

The use of image based presentation attack detection (PAD) systems has experienced an enormous growth of interest in recent years. The most accurate techniques in literature addressing this topic rely on the verification of the actual three-dimensionality of the face, which increases complexity and costs of the system. In this work, we propose an effective and low-cost face spoofing detector system to supplement a PPGI-based vital signal monitoring application. Starting from a set of automatically located facial feature points, the movement information of this set of points was obtained. Based on a time series analysis of the landmark position distances using peak descriptors and cross-correlation coefficients as classifiers in a sliding window, life signs have been exploited to develop a system being able to recognize false detections and biometric spoofs. To verify the performance, experiments on three different benchmark datasets (CASIA face anti-spoofing, MSU and IDIAP Replay-Attack databases) were made. The evaluation of the proposed low-cost approach showed good results (accuracy of ～85-95%) compared to more resource-intensive state-of-the-art methods.