Application of machine vision in image compression

A general-scene object-oriented image coder is presented which can encode colour TV signals at low bit-rate independent of content, complexity, and temporal behaviour of the TV sequence. This is a novelty compared to object-oriented coders developed by other researchers which are specialised in video-phone scenes. For the realisation of the coder´s generality, stereoscopic and motion information has been used for segmentation and ordering of the objects. This coder is a new approach to the concept of object-oriented coding because it can encode without the use of pre-defined models and without a priori knowledge of the scene content. It is also unique in its type because it utilises machine vision knowledge for object segmentation and manipulation. This work was done within the framework of RACE (Research on Advanced Communications in Europe) under contract R2045 “DISTIMA” (DIgital STereoscopic IMaging and Applications)