Enhanced face recognition using Cross Local Radon Binary Patterns

Abstract

In this paper we introduce a novel face representation method called Cross Local Binary Patterns(XLBP) to improve the robustness of face recognition for partially occluded and non-uniformly illuminated face images. In our method we use Radon transform to capture the coarse level shape information and XLBP to capture the texture information. Individual histograms computed on each sub-block of the face image and are concatenated in spatial pyramid fashion to get the complete face descriptor. Distance measures based on Pyramid Matching Kernel(PMK) is used to match these face descriptors. Experiments on FERET and YaleB databases show the significance of this method.