A Hybrid Coding Scheme for Videophone Images Based on Wireframe Modeling and Subband Coding

The paper presents a hybrid coding scheme suitable for coding of facial (videophone) image sequences at bit rates around 10 Kbits/s. The human head and shoulder part are parameterized by a wireframe model called Candide. The synthesized images are produced by a set of 18 parameters. Although the synthetic images generated by this model have subjectively better image quality than, for instance, those produced by traditional video coders (H.261/263), the geometric distortions caused by the model analysis, the motion estimation, and the texture mapping errors may be annoying. This problem is partially solved by extracting the visually important information from the prediction error image and coding them efficiently by a subband coder. Our simulation results show that the hybrid coding scheme provides an improved quality in the decoded image sequence as compared to the model based scheme [1]. 1. INTRODUCTION Coding of video images have been extensively studied in the past. The coding sche..