关键词: 图像分割, 医学图像, 均值距离, 模拟退火, 相似性

Abstract:

In the research of medical image segmentation, it is difficult to determine the number of segmentation classes. To solve the problem, a novel measurement for determining the number of classes named mean divergence function was formed according to the relation among three common means. And then an image segmentation method based on mean divergence and simulated annealing was proposed. In this method, the mean divergence function is used as an optimization object and simulated annealing is used as an optimization method to find the optimal segmentation threshold in overall search space. This overcomes the shortcomings of fuzzy C-means (FCM) clustering algorithm, such as it is hard to determine the number of classes and easy to get into a local extremum. Experimental results show that this method could automatically segment the medical image with focus, and the speed had significant improvement compared with the method based on mutual information.

Key words: image segmentation, medical image, mean divergence, simulated annealing, similarity