Recognizing Individuals from Unconstrained Facial Images

Abstract

This work makes an effort to address the problem of face recognition in unconstrained environments and presents a novel method of facial image representation based on local binary pattern (LBP). The method devises the appropriate descriptor that discriminates the facial features by filtering the LBP surface texture. The method, we name as augmented local binary pattern (A-LBP) works on the uniform and non-uniform patterns both. The non-uniform pattern is replaced with the majority voting of the uniform patterns which combines with the neighboring uniform patterns to extract pertinent information regarding the local descriptors. The recognition accuracy obtained by the proposed method is computed on Chi square and Bray Curtis dissimilarity metrics. The experimental results show that the proposed method performs better than the original LBP on publicly available face databases, AT & T-ORL, extended Yale B, Yale A and Labeled Faces in the Wild (LFW) containing unconstrained facial images.