Texture-Based Region Merging for Watershed Segmentation in Sonography

Abstract

Watershed transformation is one of the most reliable automatic and unsupervised segmentation techniques. However, the watershed transformation always produced over-segmentation for images which comprise noise and irregular textures. In this paper, we propose a new texture- based region merging method for watershed segmentation to solve the over-segmentation problem. The texture information is utilized in our merging process to control over-segmentation. The experimental results show that the proposed texture-based merging method performed well for images that are rife with noise, speckles, and textures