Face recognition based on dictionary learning and kernel sparse representation classifier

In the past few years, sparse representation classifier (SRC) has attracted great attention and widely used in human face recognition. Kernel sparse representation classifier (KSRC) based Metaface dictionary learning (MFL) is discussed in this paper. KSRC is a nonlinear extension of SRC. Through kernel trick, samples are mapped into an unknown kernel feature space first and then SRC will be used in the new high dimensional space. This nonlinear mapping can change the samples' distribution. Some samples which are linearly inseparable in the original space become linearly separable in the new space, which improve the performance of sparse reconstruction. MFL can obtain more refined dictionary to make the sample' projection more sparse. Surveys with ORL human face database and AR human face database show that the proposed method in this paper can achieve a higher recognition ratio than the original SRC method and many other classical methods.