Research of image segmentation based on watershed transformation

Abstract

The watershed transformation is a kind of powerful morphological tool for image segmentation, which can automatically segment images into a series of closed segmentation regions. Fracture image of great amount of information is an important basis for clinical diagnosis. This thesis is about the segmentation to the fracture image by dint of watershed transformation. However, fracture images are often mixed with noise and uneven distribution, direct application of watershed transformation will have a serious over segmentation. Over segmentation phenomenon can be curbed by markers-controlled watershed algorithm in some extent, but this segmentation algorithm is not universal. In this paper, we introduce the morphological direction gradient and marker selection algorithm based the on marker-controlled watershed algorithm. And then we propose an improved watershed algorithm. We evaluate the proposed algorithm by simulation experiment through MATLAB. The experiments' result shows that the improved watershed algorithm can prevent over-segmentation effectively and extract the target bone edge from fracture images accurately. © 2013 Springer-Verlag.