Improving the Quality of Very Low Bit-Rate Video by Selective Quantization of Facial Features

In this paper, we propose modifications to a H.263 codec for coding very low bit-rate head and shoulder video sequences. The important facial features (eyes, nose and mouth) are tracked from frame to frame and the knowledge of their positions is used to improve quantization in these regions of interest, compared to the other facial regions and the background, which have coarse quantization. Results are presented which show the effectiveness of the method using common head and shoulder video sequences.