Modular Sparse Representation Based Classification Approach for Robust Face Recognition

摘要

In this paper, a new sparse representation based classification technique has been proposed, named modular sparse representation based classification (MSRC). The MSRC is using a block-wise strategy (modular approach) for the sparse representation based classification. The MSRC is based on dividing the face images into non overlapping blocks, which leads to possibility of use overcomplete data dictionary without applying any of data dimensionality reduction methods. However, there will be a big challenge for the classifier to identify or select the blocks that can provide an accurate estimation of the query image. Therefore, we propose a new classification method that starts with estimation of the average sparse coding for each class then labels the query image based on maximum block-wise collaboration. Experimental results on two face recognition benchmarks demonstrate that the superiority of the proposed MSRC method for face recognition compared to a set of state-of-the-art methods.