Optimal Parallel Algorithms for Computer Vision Problems

The computational model on which the algorithms are developed is the arrays with reconfigurable optical buses (abbreviated to AROB). It integrates the advantages of both optical transmission and electronic computation. In this paper, instead of using the radix-2 system, a radix-x system can be used to represent the numbers of the intermediate results of algorithms. We first design two Oð1Þ time basic operations for computing the prefix sums of N and N 2 integers each of size Oðlog N Þ-bit on the AROB using N 1þð1=cÞ and N N 1þð1=cÞ processors, respectively, where c is a constant and c51. Currently, these are the best-known results. Then, based on these two basic operations, several Oð1Þ time parallel algorithms for computer vision problems are developed. These problems include shape moment computation, histogramming, histogram modification, Hough transform and median row. Based on the product of time and the number of processors used, all proposed algorithms are time and/or cost optimal. # 2002 Elsevier Science (USA)