一种改进分水岭乳腺肿块图像分割方法  

An Improved Watershed Segmentation Algorithm for Mammographic Masses Images

作　　者：李东红[1] 宋立新[2] 牛滨[1] 

机构地区：[1]哈尔滨理工大学测控技术与通信工程学院,黑龙江哈尔滨150080 [2]哈尔滨理工大学电气与电子工程学院,黑龙江哈尔滨150080

出　　处：《哈尔滨理工大学学报》2015年第5期25-29,共5页Journal of Harbin University of Science and Technology

基　　金：黑龙江省自然科学基金(F200912);哈尔滨创新人才基金(2010RFXXS026)

摘　　要：针对分水岭算法对噪声敏感、易产生过分割现象的问题,在图像的滤波和区域合并方法上做了改进.该算法首先对肿块图像做初步预处理,设计高斯差分滤波器,实现平滑滤波,增强图像的信噪比,并计算图像的梯度幅值;然后,依据传统分水岭变换算法进行粗分割,计算各个子区域的灰度均值并排序,依次合并灰度均值相似的区域,直到将整个肿块区域完整分割出来;最后,保留合并后灰度均值最大的肿块区域,去除灰度值较小的区域,得到最后的分割结果.实验结果表明:该算法相较于三层地形分割方法、自适应区域生长算法和二次分水岭算法,能够得到更准确的肿块边缘轮廓,误分率减少到23.07%,运行速度高.In view of the problem that the watershed algorithm is sensitive to noise and easy to produce the over -segmentation phenomenon, there is some improvement in the method of the image filtering and the region mer- ging. At first, the algorithm does preliminary pretreatment to lump images, and then designs the Gaussian differ- ence filter for smooth filtering. After that, the algorithm enhances the signal - to - noise of the image to calculate the image gradient amplitude. Then the area of the gray - scale average can be calculated and sorted by coarse seg- mentation according to the traditional watershed transformation algorithm. The gray - scale average of similar area should be combined until to complete the whole mass region segmentation. Finally, the segmentation result can be got after keeping the maximum gray - scale average mass region and removing the grey value of smaller area. The result shows that the algorithm can get more accurate mass edge profile and the error rate reduced to 23.07%. The algorithm has the high running speed comparing with the three layering algorithm, adaptive region growing segmen- tation algorithm and two watershed algorithm.

关 键 词：高斯差分滤波 分水岭变换 灰度均值 区域合并 

分 类 号：TN911.73[电子电信—通信与信息系统]