Low Rank Representation based Discriminative Multi Manifold Analysis for Low-Resolution Face Recognition

Practical face recognition algorithms occasionally faced with the problem of low-resolution profile images. Face images taken by monitoring cameras generally tend to be low-resolution(LR) with extension to unrestrained poses, noise, lighting conditions and occlusion. In this paper, we introduce a low matrix mechanism of matching occluded or inadequate characteristic profile images to a group of high-resolution(HR) profile image representations. In previous research, for matching an LR probe to a set of HR gallery images has introduced a training-based super-resolution approach which transforms LR and HR profile images into a common discriminant characteristic feature space (CDFS) for recognition. To distinguish LR images which are constrained to noise and occlusion, we present a low matrix recovery system which combines the concept of robust principal component analysis (RPCA) and coupled discriminant multi-manifold analysis (CDMMA). In RPCA, we propose to recover a low order matrix from extremely corrupted measures for better representation ability and then perform CDMMA approach in a supervised way where discriminant characteristic features for recognition increased. And then, a standard classification method is employed for final identification.