An efficient partial occluded face recognition system

This paper presents an efficient face recognition system which is invariant to partial occlusion. The proposed system makes use of Eigen faces and Gabor wavelet filters to handle occlusion. It divides each face image into a set of sub-images and makes use of Gabor filters to extract features from non-occluded sub-images. Matching between two face images is performed by comparing features of non-occluded regions of two images. The system has been tested on AR face database containing 2600 occluded images of 100 subjects. Experimental results reveal better performance compared to all well known systems which are invariant to partial occlusion.