Skip to Main Content
Given a video sequence containing face candidates, detecting faces is a challenging problem, which can be attributed to the difficulty in handling the appearance variability of the face. Based on skin color segmentation combined with the saliency model, a novel method is proposed to detect human faces in videos. Firstly, a skin color model in the YCbCr chrominance space is built to segment skin-color pixels from background in frames. Then, the candidate regions of human face can be extracted using mathematical morphological operators. Finally, a saliency model is proposed to detect eyes and mouths in face regions. Experimental results demonstrate successful face detection performance over a wide range of facial variations in color, illumination and scale in videos.