Sparse Representation for Face Recognition Based on Projective Dictionary Pair Learning

Yang Liu, Jianming Zhang, Yangchun Liu

Abstract: In order to improve the recognition rate and speed of face recognition, the paper proposes a sparse representation for face recognition algorithm based on Gabor feature and projective dictionary pair learning. Firstly, we extract the local feature of the face image at multiple scales and orientations by using Gabor transform, in that Gabor feature shows significant robustness to the variations in expression, illumination and angle. And the augmented Gabor feature reduced dimension by principle component analysis can be obtained to form a new training data instead of original training sample. Secondly, a synthesis dictionary with reconstructing capability and an analysis dictionary with the capability of quickly obtaining representation coefficients are learned jointly during the training phase. Finally, the face is identified by the reconstructed error. The experimental results on ORL, extended YaleB, AR and CMU-PIE face database show that our proposed algorithm not only get a high recognition rate but also improve recognition speed efficiently.