Human Face Detection Based On Skin Color And Template Matching

We can give a definition of face detection: Given an arbitrary image, the goal of face detection is to determine whether or not there are any faces in the image and, if present, return the image location and extent of each face. Human face examination algorithm is essential for a complete, automatic human face information analysis system.The human face detection method has been studied for more than 20 years, however, it is still a challenging task because of the variabilities from faces in scale, location, orientation(up-right, rotated), and pose(frontal, profile).This approach is developed to detecting the faces in static color images with complex scenes. It proposed a method based on skin detection, which uses the skin color and the eye features to determine the candidate faces'region, and carries on the confirmation with template matching. It is a hierarchical approach, which integrates the skin color segmentation, feature extraction and face template matching. With the elimination of false areas, the searching area will become smaller and smaller, and the detection will be accomplished eventually. Because the skin color information is easily affected by illumination, color balancing is necessary and effective. We combine the HSI color space and the YCbCr color space in order to get the region of skin color. The experiment proves that it is more accurate than the previous methods. In order to confirm the face region accurately, a template matching method which based on eye characteristic localization is used. We use an auto-adapted threshold value method to get the bi-mode image, and combine the eye template to get eyes accurately. This can reduce the number of candidate faces, thus reduces the later workload of face template matching effectively, and enhances the efficiency of this system.The experiment indicates that the skin color information is advantageous to enhance the running speed as well as reduce the inflection of complex background to face template. This approach is effective to face deflection and expression change. This system is able to achieve a correct detection rate of 86%.