Face Recognition System Based On Gabor Response Coding

As one of the most important research area of pattern recognition and artificial intelligence, face recognition is now widely used in security monitoring, financial payment, e-passports and other applications. In this thesis, we study the face recognition based on Gabor response coding and apply it to fuse gray and depth face image for person identification. Finally, we implement an embedded face recognition system.As we know, face recognition comprises two steps: features extraction and classification and Gabor wavelet feature extraction algorithm is widely used. This thesis discusses face feature based multi-scale and multi-orientation Gabor filters, then encode the Gabor Magnitude. For local enhancement, use block histogram for statistics and finally adopt LDA for dimension reduction. Experimental results on FERET database show that the Gabor surface coding algorithm has better recognition accuracy than state-of-the-art algorithms.While traditional 3D laser scanner is expensive and need complex operation, Kinect depth sensor camera is more convenient to capture both RGB and depth image. We further tested our algorithm on the publicly available Kinect FaceDB and our own Kinect FaceDB-SZU depth face Database, the fusion of gray and depth image can further improve the recognition accuracy.Finally, we build a face recognition system on embedded device. The recognition algorithm is applied to Beaglebone Black development board, where face image were captured with 850 nm near-infrared. Our system achieved 98% accuracy on the SZU_FACE_INF near infrared face database. Besides, real deployment of the system show that it is accurate, efficient and stable.