Face recognition via gradient projection for sparse representation

For face recognition, we consider the problem of automatically recognizing human faces from frontal views with varying facial expression and illumination circumstance, as well as noise. In this paper, a new algorithm is proposed, which avoids the crucial issue of feature extraction in conventional face recognition. Firstly, we use the gradient projection method to improve the performance of sparse representation classification (SRC). And then, a new algorithm dubbed classified gradient projection for sparse representation (CGPSR) is proposed, which utilizes the classification information to enhance the performance for recognizing images with noise. Simulation results demonstrate that the proposed CGPSR algorithm outperforms the previously proposed SRC-based orthogonal matching pursuit (OMP) and has a good potential in the robustness to noise.