Design and analysis of a multiprocessor for image processing

Abstract

Image processing is concerned with the manipulation of pictorial data, usually by computer. An image is generally represented as a two-dimensional array of numbers (pixels) each of which corresponds to a brightness or color value at some point in the image. There are many operations that are commonly performed on images in order to enhance them for some specific purpose: filtering, level transformations and thresholding, edge and region enhancements, feature recognition, statistical measurements, and so on. A characteristic of many of these operations is {ital shift invariance}. Operations such as edge enhancement and noise reduction are usually applied to small regions of the object image at one time. In this case, modifying one portion of the image has little or no effect on other portions some distance away. The processing of each region is independent of all of the others and can be carried out in any order. If many processors are available, then each region could be operated on concurrently with a resulting savings in the time needed. Some operations can be applied to rows or columns of an image: the Fourier transform is an example. To compute a two-dimensional Fourier transform, the transforms of the rows of themore » image are computed, and then the transforms of the resulting columns are computed. A row or column can be thought of as a rectangular region of the image, and again the regions can be processed concurrently if the processors are available. In this paper, a multiprocessor system is described that has been designed with image processing as its specific task, while attempting to minimize cost and complexity.« less