Face recognition by edge distribution

Summary form only given. Human beings have a remarkable ability to recognize human faces. Though many face recognition methods have been proposed, it is still very difficult to recognize human faces using image processing techniques. In this paper we consider that edges are an important factor for recognizing human faces. It is well known that a caricaturist represents the characteristics of a human being using only several lines. From this point of view, we have a consideration that the edge distribution on a human face becomes a good feature for face recognition. In our algorithms, a face region is segmented from the captured image in the natural office environment and then eyes and mouth are detected Based on the detected eyes and mouth, firstly the rectangular region on a face is bounded by eyes and the lower lip and the edge density of this rectangular region is normalized by the value of 20% edge density. The procedure of edge density normalization are as followings: 1) make distribution histogram of pixels using gray values, 2) determine the threshold gray value at 20% rank from the highest value, 3) select pixels which have a higher gray value than rite threshold value. Secondly, this rectangular area is divided into 10/spl times/10 cells and each cell has the edge distribution value itself. Thirdly, edge distribution of 10/spl times/10 cells are compared to that of model patterns and then the most similar model is selected as the first candidate.