Adaptive matching wavelet networks for face recognition

This paper describes a novel adaptive matching method for face recognition. We employ an face bunch graph (FBG), which has reconstructed Gabor feature on each FBG nodes. The reconstructed Gabor feature comes from the orthogonal analysis of a family of Gabor wavelet coefficients, which reduces the redundant information. After selecting fiducial points roughly by elastic bunch graph matching (EBGM) algorithm, the reconstructed feature was used for the relocation. The wavelet networks (WN) and best matching fiducial points´ location contain the discriminative information of faces. We proposed a new approach to compute the similarity between two faces on both the global means and topological means. The experimental results show that our algorithm is an effective method compared with EBGM, PCA, HMM face recognition approaches.