Multi-view face detection based on cascade classifier and skin color

In this paper, we propose a multi-view face detection method which combines Adaboost-based face detection and skin color information to improve the overall detection performance. First an input image is sent to a detector consisting of three parallel individual classifiers, and the output of each individual classifier is fused by some criterions to get a coarse merged result, and then rotate this original image by an angle of ±300 [1], so another two detection coarse results are obtained in a similar way. After these coarse results are merged into refined ones, each detected region contained in this refined result is filtered by a skin color detector, and the output of this detector is used to make up a final result contains regions supposed to have faces. Experimental results show that this combined method works well and can achieve a high detection rate of almost 93% on a dataset containing 270 faces with various poses and expressions.