DIST - University of Genoa, Italy
Image processing is one of the great challenges of computing, and image processing problems have to be solved in different application fields like telecommunications, biomedical instrumentation development, robotics and many others. In each of these fields, the image processing part is one of the most computationally heavy and time consuming parts. Very often, image processing applications need both to process enormous quantity of data and to perform complex computations. For these reasons a large number of parallel computers dedicated to image processing tasks have been developed. These machines usually consist of a set of processors homogeneous in terms of computing power and can be modeled using a single computing paradigm. Nevertheless, it is well known that image processing problems are usually structured so that no homogeneous architecture is able to perform well in every task forming a complex application, but some tasks are well suited for a computational paradigm while others are well suited for a different one. In this paper we describe the SM-IMP system, an etherogeneous, image processing oriented parallel architecture built using commercially available and highly standard computing and interconnection systems and we avaluate its performance using the standard MPEG coding algorithm as a benchmark.