Automatic face segmentation for 3-D model-based video coding

In achieving a completely automatic 3-D model-based video coding system, face segmentation plays a crucial role. The aim of this paper is to present a fast and automatic face segmentation algorithm. The face is first segmented using skin colour information. Moreover, the feasibility of the criterion of only using one chrominance plane, Cr, to segment skin is demonstrated. Finally, the hair can be removed from the segmented face region by calculating the standard deviation of statistic units. The statistic units are a set of 4 by 4 blocks containing standard deviation values of the corresponding luminance blocks.