【作者】 王静；

【导师】 王乘；

【作者基本信息】 华中科技大学 ， 系统分析与集成， 2004， 硕士

【摘要】 在数字可视人项目研究中,首先必须对原始人体彩色切片图像进行组织器官的区域分割。图像分割技术是图像处理与计算机视觉领域的前沿课题,但是已有的算法基本上都是针对灰度图像的,而对于彩色图像的分割研究还比较少。对于彩色图像的区域分割在国外主要通过半自动的分割方法实现的,而在国内,由于切片图像的质量相对较差,则主要是通过手工方法来提取器官轮廓,这对于处理海量的人体切片数据集来说显然是非常耗时耗力的。因此探索一种自动化程度较高的彩色图像的区域分割算法就显得非常有必要了。本文对目前国内外存在的纹理特征计算方法以及图像分割方法进行了较为全面的回顾与分析,根据人体彩色切片图像的特点提出了一种基于分形纹理特征和颜色信息结合的组织器官提取算法,并与其它纹理特征计算方法进行了对比。该算法将复杂的彩色图像分形维数计算转化为单色图像分形维数的计算,将纹理特征和颜色信息相结合,并且在区域生长过程中采用了多尺度区域生长准则。算法中引入了分形特征距离和颜色距离,并分别为其赋予权值a、b,通过调整权值的取值,可以将该算法应用于不同区域,有较强的灵活性。整个论文分为五个部分,第一章介绍了虚拟人研究的大环境,包括虚拟人的概念、实现过程以及国内外研究概况及其意义。第二章中系统介绍了目前存在的图像分割方法,阈值分割、区域分割、数学形态学分割以及其它。第三章中较为详细的介绍了目前已有的纹理特征计算方法,并重点介绍了我们用到的基于分形模型的方法。第四章对已有分割和纹理特征计算方法的总结与分析,提出了基于多种分形纹理特征和颜色信息结合的区域生长的分割方法,最后给出了以人体切片数据计算的结果图以及本文算法与其它算法对比的结果图。第五章总结了全文的工作,并对下一步的工作进行了展望。更多还原

【Abstract】 In the research of visual human, tissue’s region segmentation of original color human slice image is essential and first-line. Technology of image segmentation is a difficult and advanced task in the field of image processing and computer vision. Unfortunately, most of algorithms are suitable for gray image, not for color image. Segmentation of color image is mainly realized semi-automatically abroad and manually in our country. It is time-consuming and laborious obviously to deal with numbers of slice images. Therefore, it is very necessary to find an algorithm of color image segmentation having relatively high automation. In this dissertation, author reviews and analyses several kinds of methods of texture feature extraction and image segmentation respectively. On the basis of analyzing the characteristic of domestic color human slice, author proposes an algorithm of tissue segmentation that combine the fractal texture feature with color of slice image. In addition, this algorithm is compared with gray co-matrix algorithm. The algorithm proposed by author converts complicated computation of color image into the computation of homochromy image. It combines image’s texture feature with color, and uses multiple homogeneity criterions in the process of region growing. The algorithm is flexible and can be used in different regions by adjusting the weight value of fractal feature distance and color distance between two pixels. This dissertation includes five sections. In the first section, background of visual human is introduced, including the conception, the significance and developed history at home and abroad. In the second section, some image segmentation methods, such as by threshold, by region and by mathematical morphology and so on are described. In the third section, some methods of texture feature computation are set forth, especially the method based on fractal model. In the fourth section, an algorithm suitable to the color human slice is proposed and given a detailed description. Furthermore, the result <WP=6>diagrams are given at the end of this section. The whole work of this dissertation is summarized and the consequent work is proposed at last.更多还原