Kernel low-rank embedding dictionary learning for face recognition
Yuqi Pan, Mingyan Jiang · 2017
A novel face recognition algorithm - Kernel Low-rank embedding Dictionary Learning (KLEDL) is presented in this short paper. In KLEDL, we jointly realized dictionary learning and data information reduction at the same time. The core content of KLEDL is to make the threshold of coefficients distance from between-class to within-class as large as possible, for the purpose to improve the classification recognition rate and gain more discriminative features. Furthermore, the integrity of the global and local feature is preserved which is very important to sample reconstruction. As we know, sample reconstruction is proved to be fundamental for dictionary learning and low-rank representation to obtain an excellent recognition rate. Experimental results fully illustrate KLEDL outperforms all compared modern algorithms with several experiments operating in the face databases of AR, CMU-PIE and Extended Yale B (EYB).