PM4—A reconfigurable multiprocessor system for pattern recognition and image processing

Abstract

Pictorial information is often described by digitized arrays, syntactic (and semantic) strings and high-dimensional trees or graphs. The analysis and extraction of meaningful information for pictorial patterns by digital computers is called pictorial pattern analysis. Pattern analysis tasks require a wide variety of processing techniques and mathematical tools. In most machine intelligence systems, large computers are employed to process pictorial information. Because most image processing tasks require only repetitive Boolean operations or simple arithmetic operations defined over extremely large arrays of picture elements (pixels), 1 the use of large computers with rigidly structured sequential or parallel processors may result in intolerable waste of resources. 2 For example, the array-structured ILLIAC IV 3 and STARAN 4 are efficient for processing fixed-length vectors, but are very inefficient for mixed scalar and vector operations, due to the fact that multiple instruction streams do not exist simultaneously in these supercomputers.