Abstract: In view of the defect of high misdetection rate in the current Face Detection, this paper presents a new algorithm which is fusion of the skin-color model and wavelet. First, we use the skin color model to determine the potential facial position, and then separately carry on the examination to the eyes and the mouth, thus finally ensuring the real facial position. The eyes are detected based on the geometry location after the wavelet transformation, while the mouth is detected based on the Fisher classifier. This method realizes fast and accurate single face localization in the complex background. Compared with the traditional YC_bC_r skin color model, the combination of color space, the wavelet transformation, and the Fisher classification in face detection enhances the examination precision. Experimental results show that the correct detection rate is 91％ and that the error probability is only 2％.

Key words: face detection, Fisher classifier, YC_bC_r model, wavelet transform