Face Recognition: PCA vs. l1 Minimization

Abstract

Automatic Face Recognition techniques face problem when it comes to recognizing face in different expressions, poses, inappropriate lightening, occlusions and disguise. We have used PCA and l1-minimization for the recognition process and casted these problems in the experimental result thus giving a comparison between the two methods. PCA being the first ever method used for the face recognition fails when there is a occlusion or disguise. Thus we have argued that the new method based on sparse representation calculated by l1-minimization solves the problem. We have first used PCA for dimensionality reduction and then used l1-minimization for better correlation and effective result. The postulates of sparse representation predicts how to choose the training images such that the result is efficient, handles errors due to the occlusion and disguise equivalently increasing the robustness of the framework.