Coding of image sequences using a layered 2D/3D model-based coding approach

We present a layered model-based coding approach for videophone applications which employs multiple coding modes specifically optimized to improve on specific image content. The images are separated into regions, each compliant to a certain object and motion model, e.g., 2D objects with affine motion or 3D objects with rigid motion. Each region is coded separately, with an optimized coding scheme and an individual set of coding parameters. The remaining image regions for which no model can be constructed are coded with a standard MPEG-4 coder in a fallback mode. Our results show significant improvements of the subjective image quality compared to a standard H.263 coder.