Abstract
In this paper, a supervised feature extraction algorithm named collaborative representation discriminant projections (CRDP) is proposed for face recognition. CRDP explicitly takes into account the within-class neighboring information and between-class neighboring information. CRDP constructs graph and corresponding edge weights simultaneously through collaborative representation (CR). Further, by integrating graph embedding(GE) with Fisher criterion, the discriminating power of CRDP is further boosted. The experimental results on the PIE face databases show that CRDP can achieve better recognition performance.