High-performance computing and applications in image processing and computer vision

Abstract

A variety of parallel computer architectures are being used today to cope with the computationally intensive tasks in the areas of image processing and computer vision. Most image processing algorithms can readily exploit SIMD (Single Instruction, Multiple Data Stream) machine architectures. The mapping of these algorithms to such machines is rather straightforward. The fine granularity parallelism and regular data units are inherent in the nature of these algorithms. The basic disadvantage of the SIMD systems is their inadequacy to handle problems where the data involved is high level and irregular and the operations defined on them are complex. The MIMD (Multiple Instruction, Multiple Data Stream) machine architectures have the potential to deal with this kind of problem, common in computer vision. Our studies show that a reconfigurable system termed as the Reconfigurable Multi-Ring Network (RMRN) supports image processing as well as computer vision algorithms within the same architecture framework. We show the RMRN to be a viable architecture for image processing and computer vision prob- lems by demonstrating the parallel computation of a set of imaging algorithms on the SIMD, SPMD (Single Program, Multiple Data Stream), or cluster of workstations. In particular, we will address the problems of stereo image reconstruction, image classification, and image segmentation used in digital mammography.