Sparse Representation Classification Based Linear Integration of l~1-norm and l~2-norm for Robust Face Recognition

摘要

Sparse representation based classification (SRC) has been introduced as a new algorithm for face recognition classification instead of the classical gradient-based algorithms. However, there are some limitations that influence the robustness properties in SRC. One of the most effective parameters that impacts the SRC performance is the directory of training samples. It should contain enough samples to represent the test image with significant features that can achieve high classification rates. To overcome this limitation, we propose a new technique that utilizes SRC with least square and takes the advantages of regularization parameters with both l~1-norm for sparsity and l~2-norm for holding the correlation between the samples. By integrating l~1-norm and l~2-norm with regularized regression method, the SRC classifier improves the face recognition accuracy. The performance evaluation of the proposed algorithm is conducted on several publicly available databases and observed promising recognition rates compared to a set of state-of-the-art techniques.