基于快速 mean-shift聚类与标记分水岭的图像分割方法

摘要

For the problem of traditional watershed transformation algorithm that it is prone to oversegmentation in image segmentation process, we propose an image segmentation algorithm which is based on fast mean-shift clustering and marking watershed transformation. First, it uses fast mean-shift clustering to pre-process the original image and determine the segmentation areas and the number of clusters;then it employs sobel operator to carry out gradient processing followed by morphological operation on the processed image, it also distributes different marks to each water basin, and accesses every pixel in ascending order, and finally immerses them into water basin in turn to com-plete image segmentation.Experimental results indicated that the proposed method could effectively segment medical images, and could also overcome the oversegmentation issue incurred by watershed transformation.%针对传统分水岭变换算法在图像分割过程中容易产生过分割问题，提出基于快速mean-shift聚类和标记分水岭变换的图像分割算法。首先利用快速mean-shift聚类算法对原始图像进行预处理，确定分割区域和聚类数目；利用sobel算子进行梯度处理；对处理后的图像做形态学运算，并给每个集水盆分配不同的标记，按升序访问每个像素点，依次浸没到集水盆中，完成图像分割。实验结果表明，该方法可以有效分割医学影像，并解决了分水岭变换引起的过分割问题。