Image coding using an analysis-synthesis technique

Model-based coding is becoming an important technique for image data compression. One approach is specific to images containing a particular domain of objects such as a human head-and-shoulders. A computer model of a head-and-shoulders is generated by analysing a sequence of images and locating a set of important feature points. Image texture is mapped onto the model which is used to generate a sequence of moving synthetic images. Motion in the synthetic image is made to approximate the motion in the original sequence by tracking feature points. The paper describes simulation work on this type of model-based coding. It also discusses the analysis aspects of a model-based coding system for videophone images