Skip to Main Content
In this paper, a face recognition method based on edge features of facial images is proposed to solve the problem of face recognition in varying lighting. The approach is to improve edge features of human facial images in varying lighting by reducing noise. The edge features are defined as edges positioning between two large continuous surfaces in the gradient space of facial image. Noise is defined as edges positioning between two small fragmented surfaces or in shadow regions. In order to reduce noise, our proposed method utilizes a gray-level pyramid for clustering shadow regions and small fragmented regions. The mean-field method considering four neighbors' equilibrium is successively applied to reduce small fragmented areas. In this process, the noise surfaces are masked and reconstructed by the information in the symmetrical parts of facial images. Subspace method is used as a classifier. It has been proved that the proposed method can provide the accuracy of 100 percent for subsets 2, 3, 4 and 5 of the Yale facial database B when all images in subset 1 are used for training. It also provides accuracy of larger than 85 percent for those 5 subsets when only one image in subset 1 is used for training.