【作者】 陈优广；

【导师】 顾国庆；

【作者基本信息】 华东师范大学 ， 系统分析与集成， 2006， 博士

【摘要】 边界跟踪与填充是图像处理的基本问题。链码间的转换是从已知一种链码获得其他链码的便捷方法。链码是获得图像几何特征的重要手段。文档图像的倾斜校正和表格识别是字符识别技术最重要的应用领域之一。 本文从边界跟踪、链码转换、区域填充、图像几何特征的计算到基于链码的表格处理软件，对链码相关的算法和链码的应用问题进行较为宽幅度的研究。本文的工作及研究成果可以归纳为： 1、分别就八近邻图像和四近邻图像给出了边界跟踪、顶点链码抽取及围线树结构的生成算法。首先通过构造像素顶点矩阵，利用像素顶点矩阵跟踪边界、抽取边界的顶点链码并生成围线树结构。其次设计了边界跟踪自动机，利用自动机的输出获得边界的顶点链码，自动机跟踪所有图像边界的同时生成围线树结构。这两种算法都是线性的，且适用于任意复杂图像区域，生成的围线树结构是一棵以围线类为节点的双向指针树。 2、研究了正方形点阵上二值图像的几种链码之间的相互转换算法。包括Freeman缝隙码与顶点链码之间的相互转换算法，四方向Freeman链码与顶点链码之间的相互转换算法和八方向Freeman链码与顶点链码之间的相互转换算法。这样只要获得一种链码就可以得到其它的链码表示，由某种链码获得的图像信息也为其他链码所共享。 3、分析研究并发展了基于Freeman链码、缝隙码和顶点链码的区域填充算法。算法包括一种基于Freeman链码的区域填充算法、一种基于缝隙码的区域填充算法、一种基于顶点链码的区域填充算法和一种新的奇偶点配对的区域填充算法。还给出了算法的复杂度分析，并与现有的填充算法进行了实验和比较，实验结果表明这些新算法的速度优于现有算法，特别对多连通或整幅图像填充时，由于不对区域内部孔洞填充，算法运行速度有很大提高。 4、利用区域边界的顶点链码表示，给出了计算边界点坐标和边界上任意两点之间的欧氏距离的坐标标定自动机，还给出了计算图像几何矩和图像Euler数的算法。 5、给出了一种表格文档图像的倾斜校正和表格单元格的实时识别算法，在图像倾斜校正和表格单元格识别算法的基础上，给出了一个基于图像的填表系统的设计与实现方法。更多还原

【Abstract】 Boundary tracing and region filling are basic problems in digital image processing. Transforming chain code from one to another is convenient by transformation algorithms. It’s the important method to obtain geometric characteristic of image by chain code. The slant correction and form recognition of document image are the most important applications in the field of Optical Character Recognition.Research in this paper includes boundary tracing, transformation algorithms among chain code, region filling algorithm, obtaining geometric characteristic of image and form processing software based on chain code, algorithm and application of chain code are researched widely. The achievements are listed below:1. Algorithms are proposed for tracing boundary and extracting vertex chain code and obtaining the tree structure of contours from 8-adjacent and 4-adjacent images. There are two algorithms in all. Firstly, trace region boundary and obtains chain code to construct tree structure of contours through a pixel vertex matrix. Secondly, designed automatic tracing machine to obtain vertex chain code and tree structure of contours at the same time of tracing the boundary of the image. Both the algorithms are linear and can be applied to arbitrary complicated image region.2. This paper researches several transformation algorithms among different chain codes of binary image on square lattice. These algorithms are as follows: transformation algorithms of Freeman crack code and VCC, transformation algorithms of 4-direction and VCC, transformation algorithms of 8-direction and VCC. Once one kind of chain code of an image is obtained, the image information can be shared by other chain codes.3. Region filling algorithms based on chain code are analyzed and improved, including the region filling algorithms based on Freeman chain code, crack code, vertex chain code and a new parity matching algorithm. Finally, complexity analysis of each proposed algorithms are given. Experiments indicate that compared with the existing ones, these new algorithms work better. They work much better than the current methods especially in the case of multi-connected region or the whole image, because the proposed algorithms don’t fill the holes in the region.4. In the paper, a coordinate labeling automata and an algorithm computing geometric moment and Euler number of image using vertex chain code are proposed.更多还原