Detection of Video-Based Face Spoofing Using LBP and Multiscale DCT

Despite the great deal of progress during the recent years, face spoofing detection is still a focus of attention. In this paper, an effective, simple and time-saving countermeasure against video-based face spoofing attacks based on LBP (Local Binary Patterns) and multiscale DCT (Discrete Cosine Transform) is proposed. Adopted as the low-level descriptors, LBP features are used to extract spatial information in each selected frame. Next, multiscale DCT is performed along the ordinate axis of the obtained LBP features to extract spatial information. Representing both spatial and temporal information, the obtained high-level descriptors (LBP-MDCT features) are finally fed into a SVM (Support Vector Machine) classifier to determine whether the input video is a facial attack or valid access. Compared with state of the art, the excellent experimental results of the proposed method on two benchmarking datasets (Replay-Attack and CASIA-FASD dataset) have demonstrated its effectiveness.