摘要随着现代化程度的加深，传统的医学手段已经不能满足大容量的医学分析， 为了得到细胞甚至更小单元的信息，图像分割应运而生。但遗憾的是，至今没 有出现一种可以分割所有图像的算法。所以，现在的图像分割都致力于研究适 应域更为宽广的算法。根据算法使用频率的高低，本文选择代表红白细胞的近 圆形粘连细胞作为分割对象，设计基于多尺度特征的细胞分割算法。
论文的核心算法是基于距离变换的分水岭算法。 全文首先从图像预处理入手，分为图像去噪和形态学变换两个部分，其中去噪过程中选择中值滤波得到对比度强、边缘平滑的图像；形态学变换则采用 膨胀运算减去腐蚀运算效果图突出图像特征，获得满意的预处理效果图。83864
接着由控制标记法引出抑制分水岭过分割的思想，于是运用距离变换得到 细胞骨架并取其区域极小值，得到细胞中心，成功实现细胞定位后巧妙结合强 制最小技术，在分水岭分割过程中控制集水区只在中心点周围形成，获得准确 的细胞分割图像。
最后利用区域标记对细胞进行计数。 本文的创新点在于强制最小技术的引用，缺点表现在只适用于近圆形粘连
细胞，对其他特异性较强的细胞分割效果不佳。
毕业论文关键词：分水岭算法 距离变换 强制最小技术
Abstract With the deep modernization of the society, traditional medical analysis has been unable to meet the large capacity of medical tests。 In order to obtain information of cells and even smaller units, the use of image segmentation is born。 Unfortunately, so far it does not appear a way to split all images perfectly。 So, now the image segmentation is working on broader domain adaptation algorithm。 According to the use of frequency of an algorithm, this paper selects nearly circular adhensive cells which represents red or white adhensive cells in bodies blood as the segmentation object。 It designs a cell segmentation algorithm based on multi-scale features。
The core segmentation algorithm of this paper is a watershed algorithm based on distance transform。
First, it starts from the image preprocessing, including image denoising and morphological transformation, the front selects median filter to obtain high contrast, smooth-edged images; morphological transformation uses dilation pictures to substrate erosion pictures to outstand image features。 As a result, it gets satisfactory pictures。
Then the control labeling leads to the thinking of inhibiting the over-segmentation of watershed algorithm, so it uses distance transform to get the cells’ skeleton takes the minimum area to find the center of the cells。 In the other words, it gets cells’ localization successfully。
With the help of mandatory minimum technology, watershed segmentation can only produce areas around the center of the minimum; at last, it obtains accurate cell segmentation images。
Finally, the area of labeled cells was counted。
The innovation of this paper is the use of mandatory minimum technology, and the shortcoming is it only supplies to the nearly circular adhensive cells not for others。
Keywords: watershed algorithm distance transformation mandatory minimum technology
目录
第一章 绪论 。。 3
1。1 研究背景和意义 3
1。2。1 本文研究细胞图像特征 。 4
1。2。2 细胞图像分割国内外现状 4
1。3 本文结构安排 。 6
1。4 本章小结 6
第二章 图像的数学形态学变换 8 显微镜图像中的细胞分割方法研究:http://www.youerw.com/tongxin/lunwen_99103.html