Local Similarity based Linear Graph Embedding: A Robust Face Recognition Framework for SSPP problem

As very popular methods for face recognition, subspace learning algorithms have attracted more and more attentions. However, they will suffer serious performance drop or fail to work when encountering SSPP problem. In this paper, we propose a robust framework called local similarity based linear graph embedding to solve this problem. Motivated by "divide and conquer" strategy, we first divide each face image into many local blocks and classify each block, and then integrate all the classification results by voting. To classify each block, we propose local similarity assumption, which not only makes LDA feasible to SSPP problem but also improves the performance of other subspace learning methods. Finally, we further summarize a general framework to unify these local similarity based subspace learning algorithms. Experimental results on two popular databases show that our methods not only generalize well to SSPP problem but also have strong robustness to expression, illumination, occlusion and time variation.