Addressing the Computational Requirements of Image Processing with a Custom Computing Machine: An Overview1

Custom computing platforms represent a new approach to high-performance computing. By configuring processing elements and interconnections at a low level, these systems can be tailored to a wide variety tasks and yet can yield performance that rivals that of application-specific hardware. Image processing tasks are computationally demanding, and therefore represent a good test of the custom computing approach. Image processing is commonly performed either slowly on general-purpose systems or on costly, dedicated architectures. This paper describes the use of the custom computing engine to accomplish real-time or near-real-time performance on a variety of image-processing tasks. The architecture is described, along with the design methodology. To illustrate the design process, the implementation of a median filter is provided as a case study in the use of the system. Performance results for this and other tasks are provided. 1 Portions of this paper have been derived from [16]. It has since been upgraded and expanded to reflect the tutorial nature of this manuscript. INTERNATIONAL PARALLEL PROCESSING SYMPOSIUM 1995 2