Component-based multi-classifier fusion for face detection

A wide variety of present approaches work well in detecting frontal faces, but they are often unable to detect faces with partial occlusion, rotation and strong shadows. To address this problem, we propose an efficient technique. First, we filter the face-like regions from the input image using skin-color model. And then component classifiers are used to detect the faces' components from these potential regions. At last, according to geometrical relation between components, we locate faces. The experimental results illustrate that this method has superior efficiency on both of occluded face and non-occluded face.