A functional methodology for parallel image processing development

Parallel image processing has been a topic of interest for many years, but has not yet delivered a design methodology which can maintain portability across rapidly changing parallel implementation technologies. The two hurdles to be overcome are the difficulties of re-arranging application code into a parallel form and the lack of a general and universally supported parallel machine model. This paper addresses these problems using a functional language transformation, illustrated in the context of a particular application. We show that the overhead introduced by the transformation is relatively small, but the benefit derived is substantial, since the functional programming discipline enforces an implementation-independent definition of core parallel requirements, which can then be mapped onto a broad set of parallel architectures, ranging from shared and distributed memory conventional multiprocessors to direct hardware implementations constructed using silicon compilers and FPGA technology.