3D Face Recognition Based On Collaborative Representation Classification

Face recognition is a biometric technology, which has become potential field and broad application. It is friendly, non-contact and non-disturbing. The current face recognition is mainly based on 2D intensity image. Although the technology of 2D face recognition has become more and more mature, it is still sensitive to variations in illumination, facial posture and expression.2D image is only contracted plane projection of 3D face.3D depth map data which is not sensitive to light changes and contains the inherent information. Taking full advantage of depth information of 3D face is in general able to overcome the problems resulting from illumination, pose and expression variations in 2D face recognition. Collaborative representation with fewer features can effectively express the important information of 3D face. So our thesis based on 3D face depth map, a fused residual algorithm based on collaborative is proposed. The main contributions and innovations are summarized as follows:1. We firstly overview the background, significance and the current situation of face recognition. It gives a classified summary of existing methods of face recognition, and describes several representative feature extraction algorithms in detail. Sparse representation classification and collaborative representation classification is studied roundly application in face recognition.2. To overcome the crucial problem of expression, illumination and pose variations in 2D face recognition, this approach uses Gabor feature and face intrinsic information of Geodesic feature instead of the original global feature from 3D depth map, combined with collaborative representation classification algorithm to fulfill face recognition.3. To highlight all kinds of samples of collaboration and add more discriminant information, extracts features from 3D face depth images, then fuses two features via collaborative representation algorithm, and the fused residuals serve as the ultimate difference metric. Experiments conducted on CIS and Texas face databases, the recognitions rates of the proposed method are up to 94.545% and 99.286%, and improved recognition performance under the variations of illumination, expression and gesture. Experimental results verify that the proposed algorithm is effective and robust.