Object assisted video coding for video conferencing system

An object-based video coding for video conferencing system is proposed. There are two main processes: segmentation process and face detection process. The segmentation process is used to segment each frame of a video sequence into two non-overlapping regions, namely foreground and background. A novel face detection technique based on chrominance and the contour of the segmented region is applied to the foreground region. Smaller quantization step is used for the facial region to improve viewer's perception while a larger quantization step is used for the background to compensate the coding efficiency. The remaining regions are kept in normal coding quality to prevent degradation of important information other than the facial regio.