Adaptive search and image processing with transputer arrays

Abstract

Transputer arrays and the Trollius operating system are explored as means for implementing adaptive search experiments. Synthesis of image operators with evolutionary learning is a search process involving a large population of candidate solutions which undergo systematic variation and evaluation. A genetic algorithm produces the candidates with the highest performance measures (survival functions). These solutions have the highest probability of reproducing and increasing their attributes in the population. It is noted that such search processes have a natural parallelism which facilitates the use of transputer arrays. It is concluded that the introduction of transputers provides a relatively inexpensive means of enhancing the computational power necessary for attacking practical problems in computer vision and evolutionary search. A secondary advantage is that they provide a flexible testbed to stimulate further investigation and evaluation of parallel algorithms for distributed computing systems.<>