Joint dimensionality reduction for face recognition based on D-KSVD
Yong-Sheng Hu, Jie Li, Yang Zhao, Jun‐Guo Lu, Jianbo Su · 2016
Face recognition based on sparse representation is investigated in this paper. Dimensionality reduction is the process of projecting original image data into a low dimensional space, and usually be conducted before dictionary learning. However dimensionality reduction may lose information that is important to the face recognition task. Although the accuracy rate of face recognition varies with different projection matrixs, most of the existing methods ignore the importance of dimensionality reduction. To exploit more information from raw data, a new algorithm that combine the dimensionality reduction and dictionary learning based on discriminative K-SVD(D-KSVD) is proposed. The algorithm conducts dimensionality reduction and D-KSVD jointly for the purpose of learning a more powerful dictionary that has less reconstruction error. Experiments on several databases including AR, Extended YaleB and CMU PIE show the effectiveness of the proposed algorithm.