Compressed sensing is a signal processing technique.udThe entity signal can be efficiently reconstructed if the sparse representation is determined. The sparse representations of all the test images are determined with respect to the training set by computing the l1-minimization. However, sparse representation which involves high dimensional feature vector is computationally expensive. Thus, discriminative features that could perform accurately for the face recognition system under visual variations, such as illumination, expression and occlusion have to be selected carefully. In this paper, feature selection method in the application of face recognition based on sparse representation classifier (SRC) is proposed. The proposed technique first divides the images of a few subjects into chunks. Then, it selects the feature subsets based on distance based measurement, the residual, and recognition performance, the accuracy. Extensive experiments with visual variations are carried out by using ORL, AR and Yale databases.
展开▼
