Implémentation of image processing algorithmes ou a multi-transputer system

Voir Plus Voir Moins

Afficher le Résumé

This work fonns a part of a project which is concerned with theinvestigation of the feasibility of using transputers and occam to builda nigh performance concurrent image processing system. In particular,this thesis is concerned with the implementation of a selection of lowlevel image processing operations on a multi—transputer system. A setof commonly used low level hnage processing algorithms have beenhnplemented on a transputer network utilising the image parallelism technique to exploit the concurrency in the algorithms. This involveseach processor performing the whole algorithm on assigned sub—images.There are two ways of allocating sub—Dmages to processors:statically and dynamically. In the static load allocation method, theimage is partitioned into a number of sub—images which are pre—allocated(before run—time) to different processing elements to be processed inparallel. in the dynamic load allocation scheme, sub—images areallocated dynamically (at run—time) to available processors, whichprocess them, send results, and then request for more work. Both thesetwo aproaches have been investigated during the course of thisinvestigation.Results have shown that similar eri onnance is obtained for thestatic allocation scheme and the dynamic allocation scheme whenselecting the optimm load—size. A multi—transputer system consistingof 42 processors was found to hnplenent the algorithms between lO to l3times faster than a single transputer system. The inability to achieve