Window theory of local image processing (IP): a framework for mapping local IP algorithms onto local IP systems under real time constraints

This paper introduces a window theory of local image processing (IP). An associated framework for mapping local IP algorithms onto local IP systems is presented. The window theory relies on two fundamental concepts: image windows and window operators. The starting point of the mapping methodology is thereby the local IP algorithm specifications in the form of instruction sequences referring to window operators as macro-instructions. The target system architecture is a local IP system architecture, consisting of a POM-processor array as coprocessor and a VLIW-processor as POM-Front-End. The main features of this mapping framework can be summarized as follows: optimality through execution time minimization of the given local IP algorithm, tractability of the underlying ILP-model, globality through quantification of all hardware resource requirements and real-time compatibility.