| 摘要 |
In a method of coding a video signal for low bit rate coding applications such as video teleconferencing or telephony, an encoder comprises an automatic face location detection method which models face contours as ellipses and transmits the face location model parameters to the decoder. This face location information may be exploited with use of at least two techniques, each in accordance with another illustrative embodiment of the present invention. In one technique, referred to herein as "model-assisted dynamic bit allocation," a three-dimensional subband-based coding method is enhanced by providing two quantizers per subband -- a fine quantizer which is used to code pixel data inside the detected face location model, and a coarse quantizer which is used to code pixel data outside this region. Thus, the coding quality inside the facial regions is improved relative to the coding quality of the remainder of the image. In another technique, referred to herein as "model-assisted motion compensation," a motion-compensated coding method is enhanced by automatically computing motion vectors for pixels inside the face region based on the relative positions of detected facial models in successive frames. No motion information needs to be explicitly transmitted to the decoder, since the motion vector may be recomputed at the decoder. |
