Content–based Coding of Videophone Sequences Using Automatic Face Detection

A content–based coding scheme for the transmission of videophone sequences at very low bit rates conforming to the MPEG–4 standard is presented. The goal is to improve the image quality in the facial area of a person, at the expense of a lower quality in the remaining image, which is subjectively less important for the communication partner. In a first step, the face of the talking person is detected automatically. Then, each image is coded and transmitted as two different video object planes (VOP): the face VOP is formed by the facial area, the residual VOP by the remaining image. Thus, a large amount of the available bit rate can be used for coding the face VOP in good quality, while the residual VOP is coded at a lower quality. Using typical videophone sequences and compared to a standard scheme which codes the whole image at the same quality, the proposed scheme shows significant improvements of the image quality in the facial area.