Face recognition based on deep feature mapping to kernel space
Meng Wang, Qingqing Liu, Xiaotong Cheng · 2022
Deep face recognition has attracted more and more researchers' attention due to its layer-by-layer feature learning. However, the classical sparse representation methods have strong classification ability. Considing above two advantages of deep learning and sparse representation, this paper proposed a face recognition algorithm based on deep local dictionary (DLD) feature optimization combining weighted and joint kernel collaborative representation (WJKCR) classification. The main contribution of this paper as follows: (1) the normal probability density function is used to extract local blocks from multi-angle face images. (2) Deep CNN and transfer learning theory were used to extract features from image blocks, and the feature set was selected as the optimal local feature dictionary. (3) Use the weighted and joint kernel collaborative representation for face classification to verify the effectiveness of our algorithm.