Region-of-interest video coding based on face detection

The ability to give higher priority to Region-of-Interest (ROI) is the emerging functionality for nowadays video coding. A simple and fast method of face detection is proposed to dynamically define ROI in real time application. We use the color information Cr and RGB variance to determine the skin-color pixels. We don't need extra preprocessing, because these two color spaces are used in most hardware and video codec standards. Then, we use low-pass filters for background to reduce used bits. For video coding system, a region-based video codec based on the H.263+ with the option mode of modified quantization is set up. We adjust the distortion weight parameter and variance at macroblock layer to control the qualities at different regions. From experimental results, the proposed method can significantly improve quality at ROI. Our method is suitable for real time videoconferencing.