Implementation of a Region Growing Algorithm on Multicomputers: Analysis of the Work Load Balance

This paper discusses and evaluates different parallel implementations of a Region Growing algorithm based on the Split-and-Merge approach proposed by Horowitz and Pavlidis [1]. Region growing is a general technique for image segmentation. The basic scheme consists of joining adjacent pixels to form regions; adjacent regions are then merged to obtain larger regions. The solution has been conceived for a multiprocessor using the SPMD (Single Program Multiple Data) programming model and executions have been carried out on a Cray-T3E system. From a parallel point of view the region growing problem is an irregular and dynamic problem which exhibits unpredictable load fluctuations. Therefore it requires the use of load balancing schemes to achieve efficient parallel solutions. In this sense, we propose both a static and a dynamic load balancing strategy. The efficiency of the algorithm also depends on the procedure applied for determining the region identifiers ( ) of the set of regions. To show this fact, we propose and analyze several strategies for the selection of the and its influence on the execution time and the load distribution. keywords: parallel algorithms, load balancing, segmentation, region growing, multicomputers.