Color image face detection: An algorithm

This paper puts forward a solution for real-time color-imaged human face detection. The color face image in RGB space is first transformed into human face color models under r,g,b chrominance which can be used to differentiate skin pixels from non-skin pixels. The color image is then divided into grid units which are arranged as a rectangle, and the ratio of skin pixels to all pixels in the grid unit is computed and the grid unit is regarded as a skin unit the ratio of which is greater than a threshold. The adjoining skin units are concatenated. The resulting area is labeled as a candidate face if its shape is quasi-ellipse or quasi-rectangle complying with proper ratios. Otherwise the resulting area should be disregarded. Finally the gray image of face area is matched to the face template to discern the real face. This algorithm overcomes the influence of complex background upon face detection and obtains higher accuracy of detection. In particular, the time consumed by this algorithm is much less than the traditional template matching method. Under a resolution of 320*240 it can reach 10 frames per second and is suitable for real-time face detection system.