Extraction of face parts by using flesh color and boundary images

Abstract

The new method for detection of the face direction is proposed. In this study, the face region was extracted according to the histogram which shows the distribution of the flesh color and histogram of edge image got by the Sobel filter. And, the direction of the face was roughly specified from the histogram which showed the distribution of the low bright pixels. Then, face parts detection was carried out using the template matching. Since rough face direction and position of parts have been specified firstly, the template matching was efficiently carried out. The result of the rough face direction was generally good, and the face parts detection such as eyes and mouth was also good. But the face parts detection in left or right direction was not as accurate as the front face.