Image Segmentation Using Dynamic Region Merging

Abstract

In region merging the there are two essential issues first is order of merging and second one is stopping criterion. This work addresses two issues which are solved by Dynamic region merging algorithm which is defined by SPRT and the minimal cost criterion. The process is start from an oversegmented image, then neighboring regions are progressively merged if there is an evidence for merging. The final result is based on the observed image. This algorithm also satisfies the certain global properties of segmentation. In this algorithm region merging process become faster due to nearest neighbor graph in each iteration. The performance of dynamic region merging algorithm is shown on natural images.