Class Specific Dictionary Learning - Local Kernel Collaborative Representation Classification for Face Recognition

Illumination, expression and occlusion are inherent problems in face recognition. Thus, this paper proposes a new method based on kernel function and segmentation. A face image is firstly blocked, and each block is mapped to a higher dimension space by Gaussian kernel. Then, combining with class specific dictionary learning, the reconstruction error corresponding different class of each block based on local kernel collaborative representation is computed. Finally, according to the reciprocal of the reconstruction error, the process from local discrimination to the global classification is completed by the form of voting. Experimental results on three face databases (Extend Yale B, AR and CMU PIE) and the mixed face database (including AR, Extend Yale B and CMU PIE) show that the proposed method has high recognition accuracy of 99.8%, 98.8%, 93.9%, 87.1 %, respectively, compared with the recent CSDL-CRC method, increased by 10.4%, 7.5%, 4.6%, 8.2%.