High Speed Image Rotation in Embedded Systems

Abstract The design of an enhanced SIMD processor, named the SIMDDE, is evaluated. The architecture has circular data paths, as well as row and column paths. The SIMDDE architecture is intended primarily for use in image recognition in embedded systems. The designed processor also might be realized with other architectures provided they meet the spatial constraints. Performance comparisons are made with a von Neumann, an unenhanced SIMD, two pipeline architectures, and a paracomputer. The methodology employs an image recognition algorithm introduced by Castleman (Digital Image Processing, Prentice-Hall, Englewood Cliffs, NJ, 1979) and an extension of the published analysis of Cantoni, Guerra, and Levialdi (Computing Structures for Image Processing (M. Duff, Ed.), pp. 43-56, Academic Press, New York, 1983). It is shown that this augmentation can increase the performance threefold at only a 6% increase in the cost as measured by the number of paths required. Comparisons of the design as realized with other architectures indicate when these might be preferred.