《粒计算在图像分割中的应用》由会员分享，可在线阅读，更多相关《粒计算在图像分割中的应用（97页珍藏版）》请在金锄头文库上搜索。
1、太原理工大学 硕士学位论文 粒计算在图像分割中的应用 姓名：李恩群 申请学位级别：硕士 专业：控制理论与控制工程 指导教师：谢克明 20090501 太原理工大学硕士研究生学位论文 I 粒计算在图像分割中的应用 摘 要 粒计算是信息处理的一种新的概念和计算范式，覆盖了所有和粒度相 关的理论、方法、技术和工具的信息，主要用于描述和处理不确定的、模 糊的、不完整的和海量的信息以及提供一种基于粒和粒间关系的问题求解 方法。粒计算的基本思想是在问题求解中粒的使用，它通过粒对现实问题 的抽象、粒之间的关系、粒的分解和合成以及粒或者粒集之间的转换来描 述和解决问题。 粒计算，通俗地讲，就是以粒为单位来进行
2、计算。粒是我们对现实的 抽象，它的目标是建立高效的以用户为中心的对于外界世界的观点，从而 支持和帮助我们对周围物理和虚拟世界的感知。人类具有根据具体的任务 特性对相关数据和知识抽象或者泛化成不同程度、不同大小的粒的能力， 以及进一步根据这些粒和粒之间的关系进行求解的能力。 基于这样的思想，本文建立了粒度分层模型，通过构建图像上的粒度 分层结构，从图像中抽取其特征，在此基础上，用这些特征进行图像分割。 该模型是基于相关关系构建的粒计算模型，它由四个部分组成：对象集系 统，相关关系系统，转换函数和嵌套系统，主要特点在于对粒的定义以及 通过粒度的层次嵌套结构进行问题求解的方法。在本部分工作中，主要研
3、 究了粒的定义、关系及其合成和分解技术，以及粒度分层模型、粒度分层 太原理工大学硕士研究生学位论文 II 结构的构建方法和模型的主要特点等；建立自己的粒度分层结构，对图像 进行分割。通过实验表明，粒度分层模型在图像分割上具有良好的应用效 果。 本文研究实现有关图像分割的算法，最主要是区域分割算法；并采用 灰度特征描述刻划图像区域；本文主要采用的是区域生长算法，对种子点 的选取，生长的法则，以及生长的方向等都进行了详细研究，并对此算法 进行了改进，成功地实现了对灰度图像进行基于子块的区域生长法。实验 证明，基于子块的图像分割算法的分割结果符合人的主观感知，分割效果 令人满意。基于子块的区域生长法
4、，充分研究了区域生长法的基本要素： 区域的数目，各区域的生长核心（种子） ；区域间相区别的性质特征，由此 构造同质判据。 关键词：粒度分层模型，粒度分层结构，粒计算，图像分割，区域生长法 太原理工大学硕士研究生学位论文 III GRANULAR COMPUTING IN THE APPLICATION OF IMAGE SEGMENTATION ABSTRACT Granular computing, which is one new concept and the computation model of the information processing, has covered all
5、 correlation theories, the method, technical and tools information about the granularity. It mainly is used in describing and processing indefinite, fuzzy, incomplete and magnanimous information, and also provides one kind question solution method based on the granules and granules relations. The ba
6、sic philosophy of the granular computing is granular usage in the question solution. It describes and solves the question through the granule to abstract the realistic question, the relations between granules, granules decomposition and the synthesis and the transformation between granules or the gr
7、anules collection. Popularly speaking, granular computing is carrying on the computation in the granule as the unit. The granule is what we abstract the reality as; its goal is to establish highly effective viewpoint which take the user as the center about the outside world, so as to support and hel
8、p us to feel the periphery physics and the hypothesized world. The humanity has the ability of abstracting or pan-turning the correlation data and the knowledge into the varying degree, 太原理工大学硕士研究生学位论文 IV different size granule based on the concrete duty characteristic, as well as further carrying o
9、n the solution according to these granules and the relations between granules. Based on such thought, this paper has established the granularity hierarchical model, through constructing images granularity hierarchical structure, we extract its characteristic from the images, and based on that, we us
10、e these characteristics to carry on the image segmentation. The model is constructed as granular computing model based on the correlation, and it is composed of four parts: The object collection system, the correlation system, the transfer function and the nesting system. The main feature lies in gr
11、anules definition and carries on the question solution method through the granularity hierarchical nesting structure. In the part, we have mainly studied the definition, the relations and the synthesis and the decomposition technology about the granules, and granularity hierarchical model, the const
12、ruction method of the granularity hierarchical structure and main feature of the model and so on; Establishing own granularity hierarchical structure carries on the image segmentation. Through the experiment, it has been indicated that the granularity hierarchical model has the good application effe
13、ct in the image segmentation. We have studied and realized the related image segmentation algorithm, mainly the region segmentation algorithm; And uses the gray characteristic description to portray the characteristic of image region; What the paper mainly uses is the regional growth algorithm, in d
14、etail researches on a seed selection, 太原理工大学硕士研究生学位论文 V the growth principle and the growth direction and so on, and has made the improvement regarding the algorithm, so as to successfully realize regional growth algorithm based on the sub block to the gray image. The experiments have proved that th
15、e segmentation results of the image segmentation algorithm based on the sub blocks conform to persons subjective sensation, and the segmentation effect is satisfied. Regional growth algorithm Based on the sub block, has fully studied the basic elements of the region growing algorithm: Region number,
16、 growth core (seed) of various regions; the nature characteristic between regions constructs the homogeneity criterion. KEY WORDS: granularity hierarchical model, granularity hierarchical structure, granular computing, image segmentation, regional growth algorithm 太原理工大学硕士研究生学位论文 IX 图索引 图 1-1 论文的组织结构 20 Fig.1-1 paper organizational structure 20 图 2-1 示意图.24 Fig.2-1 schematic drawing24 图 2-2 基于粒度分层模型的图像分割框架图.33 Fig.2-2 image segmentation frame chart based on granularity hierarchical mod
