Face verification system based on texture feature with single sample per person
Xiongjing Wang, Wenming Yang, Qingmin Liao · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2012
This paper introduces a face verification system implementation. In feature extraction, the algorithm is based on a classical texture descriptor, Local Binary Patterns (LBP). In decision making, a new method is proposed to determine the Client Dependent threshold (CD-th). Compared with the traditional fixed threshold, it significantly reduces error rate. Moreover, a symmetry factor is defined to increase frontal face detection rate. And a storage mode is designed to reduce time consumption in feature extraction. The implemented face verification system requires only one sample per person, and overcomes the difficulties appearing in multi-sample face verification system, including image capture problem, storage limitation and time-consumption. The experiments demonstrate the effectiveness of our proposed system.