Detection of Face Spoofing using Multiple Texture Descriptors
Shanmukhappa A. Angadi, Vishwanath C. Kagawade · 2018 International Conference on Computational Techniques, Electronics and Mechanical Systems (CTEMS) · 2018
Automatic face recognition systems are being increasingly developed and deployed in various applications in both civilian and defense sectors. But the emergence of techniques for producing fake face images presents a new challenge and demonstrates high security risks on such systems thus necessitating imposter detection system. Among the numerous imposter detection methods reported in literature, LBP is often used as one of the best local pattern descriptor, as local descriptors are frequently used for imposter detection but the traditional LBP descriptor has the limitation of having small spatial support area and being sensitive to illumination changes. To handle such issues of LBP based descriptors, the proposed work explored a new approach to estimate rotation invariant uniform LBP pattern features from histograms of noninvariant LBP patterns. The proposed approach makes use of both structural information and the magnitude information of 3×3 neighborhood of each pixel with center gray level value to attain more discriminative power, which is more effective for detection of face spoofing. The proposed technique has achieved efficacy 98.97% recognition rate on NUAA dataset