Several approaches to implement the merging step of the split and merge region segmentation

Abstract

The purpose of this paper is to propose several approaches for the implementation of the merging step of split and merge region segmentation. The splitting step has already been studied and its parallelization has subsequently been implemented on a transputer network. First, the most widely known merging step is described. Then, two approaches which are better suited to a parallelization are presented. Next we discuss the principle behind these approaches. Finally region segmentation according to motionbased criteria has been chosen in order to provide results to evaluate the performance of each approach. We emphasize that the description is general and can be applied to all split and merge algorithms. Therefore this work is a contributory factor in the evolution of region segmentation towards its parallelization.