Recognition of Partially Occluded Face using Block Based Mean Weighted Local Binary Pattern Feature and Adaptive Sparse Classifier

Abstract

In recent literature related to the problem of face recognition under partial occlusion is a big challenge for the researchers who are working in this domain. The accuracy of any face recognition system under partial affected parts of the face image needs to be improved. To shed some light in this area this paper addresses the face recognition problem partial occlusion due to scarf and sunglasses. Initially, the face image is divided in to the part that is affected and the one that is not affected , the non occluded portion of the face image is only used for face recognition. For detecting the occluded area the given image is separated  in to many small size sub blocks and then each block is checked for occluded or affected part  using Fuzzy Segmentation. The novel feature extraction technique i.e. block based mean weighted local binary pattern feature technique extracts features from non occluded portion and are given to the adaptive sparse classifier in order to recognize the image. The method is implemented on benchmark datasets and the results were  promising

Index Terms: Face Recognition, Occlusion, Fuzzy Segmentation, SVM, IOT, Mean weighted LBP