Weighted non-negative sparse low-rank representation classification

Abstract

In the calculation of rank minimization, the non-negative sparse low-rank representation classification (NSLRRC) regularizes nuclear norm's each singular value equally, but this limits its flexibility and ability to solve many practical problems, where the singular values with clear physical meanings ought to be treated differently. In this paper, a weighted non-negative sparse low-rank representation classification method (WNSLRRC) is proposed for robust face recognition. Our method adaptively assigns weights, which provides additional discriminating ability to the original non-negative sparse low-rank models for improved performance, on different singular values. Our method is able to assess the test sample and correct classification based on class-specific reconstruction residuals. Experimental results on public face databases testify the robustness and effectiveness of our method in face recognition. Those also show that our method outperforms other state-of-the-art methods.