Face Detection Based on YCbCr Gaussian Model and KL Transform

This paper puts a skin color model which is based on YCbCr Gauss model and KL Transform carried on face detection. The region model and the simple Gauss model of the skin color are established in the KL color space and the YCbCr color space according to clustering. Skin regions are separated from non-skin regions by the optimal threshold which is obtained by an adaptive algorithm. Then the segmentation result of the region model is used to eliminate the influence of the likely-skin in the Gauss-likelihood image. Afterwards, the binary image is dealt with morphology processing to eliminate the noise. Finally, in order to locate the face, the obtained regions of skin color are screened out with simple detection algorithm based on prior knowledge. The experiment indicates that the proposed algorithm performs well in face detection in complex background, many faces or profile. The algorithm is of fast response, high accuracy and adapts to video surveillance.