| 摘要 |
Video signals are coded (102) at a first relatively low frame rate. One or more portions of the image area, for example the mouth of a person speaker; recognised by feature extractor (106), are coded (110) at a higher frame rate. Preferably the second coding operates by resolving the pixel values of the image pertion - considered as a vector - relative to a set of orthogonal reference vectors (stored in 112) to produce coefficients which are then transmitted. Preferably, the feature extractor (106) normalises the image portion as regards scaling, position and rotation and sends data on these parameters as side information. <IMAGE> |
