A novel approach to face recognition using freeman chain code and nearest neighbor classifier

This paper proposes a novel approach to face recognition using freeman chain code as a feature extractor for face representation, and the nearest neighbor classifier for face matching. The face description stage starts with resizing, and then dividing the face image into non-overlapping sub-regions, then a set of chains (words) are extracted from each region, and assigned later into the nearest word in a Dictionary of Visual Words (DoVW). As a result, each patch is represented by a histogram of visual words. Finally, the histograms are assembled into one to describe the face image. Unlike the most of the existing methods, which require a mask around the treated pixel, our methodology depends on directional changes from the starting pixel, which allow us to obtain information on the local and also the more global structures. The face matching is performed by using the nearest neighbor classifier with Hellinger, Cosine, or Chi-square as the distances measure between histograms. Experimental results show which metrics perform well and demonstrate the efficiency of the proposed approach in terms of recognition rate compared to the other face recognition methods.