Face detection using skin color and Locally Adaptive Regression Kernels

In this paper, a new face detection method is proposed based on skin color and an image feature called Locally Adaptive Regression Kernels (LARK). A novel preprocessing is applied in this method, which includes skin segmentation and the estimation of the scale and rotation. To segment the skins from the background, a compound color space called H-CgCr is proposed based on both HSV and YCgCr color spaces, and some invalid regions are removed according to the facial structure knowledge. Then the skin pixel moments are used to estimate the scale and rotation. After the preprocessing, the LARK feature is used to locate the face in the image. A series of experiments are designed to verify the adaptability of the proposed method to scales and rotations, and compare with some other methods using part of FBBD dataset. Experimental results show that the proposed method can obtain better performance than that just based on LARK. The results show that the proposed method is effective to detect faces based on only one sample.