4 - Formalisation en occam et simulation sur un réseau de Transputers d'un processeur systolique de traitement d'images

The most significant drawback in using neighbourhood processing (linear filtering, morphology) is the cost of ils implementation, particularly in the area of image processing where the cost grows in proportion to the kernel size squared . However, in PIR [1] it was shown that a large amount of classical algorithms could be described as a cascade of peculiar elementary tells. Being highly integrable, these tells brought about the design of a neighbourhood systolic processor. Converting a functional description (e .g . flow graph, mathematical model) into a hardware structure raises several technical problems . In addition, the cost of designing an integrated circuit requires special cure, making it necessary to perform many simulations . The first part of this paper gives a short description of the decomposition method. It then goes on ta describe the power of occam as a design formalism, applying it to the cascaded tells modelisation . Furthermore, the obtained model can be compiled into a code directly loadable into a Transputer network, resulting in a performant simulation . The hierarchy of occam allows a description at different levels of abstraction, in order to fit the precision of the model to the desired type of simulation .