Parallelism for imaging applications

Numerous image processing functions involve repetitive operations and therefore can benefit from parallel processing, where performance may be significantly improved as a function of the number of processors applied to the task. One such application that requires processing to be as near to real-time as possible is vision processing and, in particular, low level vision processing. A system developed by the National Bureau of Standards (now known as the National Institute of Standards and Technology) and commercialized by Aspex Incorporated typifies the performance improvements which may be realized in vision processing through the use of parallelism. The Pipelined Image Processing Engine (PIPE) is a coarse-grained parallel system ranging from three to eight stages. Using an innovative implementation of parallel hardware and software, the efficiency in executing complex imaging algorithms may be greatly improved. As will be seen, it is extremely useful in accelerating image processing functions in the areas of point processing, spatial processing, sequence processing and Boolean processing. These functions may be used for real-time or near real-time edge extraction, motion detection and disparity map generation. Applications include inspection in food processing and electronics and textiles manufacture; robot and autonomous vehicle guidance; and research into image understanding, image processing and vision processing