Robust face recognition under illumination variation and occlusion (in english)

摘要

Face recognition in real scenarios is mainly affected by illumination variation and occlusion, and therefore in order to develop a robust face recognition system these issues should be handled simultaneously. To this aim, the steps involved in the presented framework are (i) computationally simple and efficient preprocessing chain that eliminates major effects of illumination variation and noise while still preserving the essential appearance details that are needed for recognition (ii) robust feature extraction based on Robust Sparse Principal Component Analysis (RSPCA) and Linear Discriminant Analysis (LDA) in order to deal with outliers typically effecting sample images due to pixels that are corrupted by noise or occlusion and finally (iii) a computationally efficient cosine distance based classifier. Experimental results on standard face databases show that the proposed approach is robust to large illumination changes as well as to occlusions and superior to distinguished methods in the literature.