Improved watershed segmentation with optimal scale based on ordered dither halftone and mutual information

Considering traditional watershed segmentation has a serious over-segmentation problem and marker-based watershed segmentation has a great difficulty in marker extraction, we proposed the improved watershed segmentation with optimal scale based on ordered dither halftone and mutual information. We made some improvements on marker-based watershed segmentation. Firstly, according to vision characteristics of human eye, we proposed a new marker-extraction method based on ordered dither halftone. By using Bayer ordered dither algorithm, we obtained the dithering image of original image, which contained most of structure information without the disturbance of noises, and extracted representative points as markers effectively. Secondly, we designed a new index based on mutual information to decide the optimal scale of image segmentation, which we called the relative mutual information entropy index(RMIE), and used it to decide optimal segmentation scale from multi-scale segmentation results. We make a conclusion that the segmentation result with optimal scale has the greatest RMIE, compared with all other segmentation results. From the experiment results, the proposed method can produce optimal scale segmentation result with meaningful, separate and homogeneous regions, which satisfies the human eye.