A Novel Method for Face Detection across Illumination Changes

In this paper, a novel method is proposed for human face detection. It contains three parts: illumination compensation, skin color model and template matching. First, we remove the color offset, high light and shadows that exist in human face images, and Gaussian model in YCrCb skin color space is chosen. Then, the facial skin region is marked and segmented using an improved segmentation algorithm. The features such as holes, deflection angle, center of mass, area and aspect ratio are used to make further verification Finally a template matching approach is introduced to face candidate region and then the human faces are fixed by rectangles. The proposed method has been evaluated from variation of illumination, pose and expression. Experimental results demonstrate that the detection rate improved to 84% and also indicate that our face defection approach under different illumination conditions is effective and robust