【作者】 赵娜；

【导师】 王希常；

【作者基本信息】 山东师范大学 ， 管理科学与工程， 2010， 硕士

【摘要】 随着计算机技术的不断发展,计算机的存储能力得到大幅提高,越来越多的纸质文档通过各种数码输入设备以图像的形式存储到计算机中,提供给后续处理系统应用。能够将存储到计算机中的文档图像转化为可检索、分析、重新应用的电子文档的图像分析技术已经引起了广泛的关注。从文档图像处理过渡到图像分析的一个关键步骤是文档图像分割技术,它在图像工程中占据着重要的位置。文档图像分割是图像处理领域中的一个基本而又非常重要的环节,有着重大的现实意义和理论价值,也是计算机视觉领域中的一个重要步骤。文档图像的分割效果直接影响着后续分类和识别的准确性。虽然目前针对印刷体文档图像人们提出了各种各样的分割算法,其中有些算法也取得了理想的效果,然而对于手写体文档图像的分割还是存在某些重要的问题有待解决。当前在手写体文档的分割算法中,有的算法是基于倾斜矫正之后的图像的,有的算法是只是针对特定语言实施分割的,缺乏对各种图像进行统一分割的通用算法,而且大部分的文档图像分割算法对于手写体字符拓扑结构变化比较敏感,而目前主要应用于医学图像的几何主动轮廓模型对拓扑结构的变化处理非常自然。主动轮廓模型充分利用高层信息,是一种自上而下的处理模型。主动轮廓模型为轮廓提取、立体匹配和目标跟踪等一系列视觉问题提供了一个统一的理论基础,并已经在人机交互、图像对准、目标跟踪等众多领域获得了广泛的应用。基于Mumford-Shah模型的水平集图像分割方法是一种很优秀的图像分割方法。由于Mumford-Shah模型的水平集图像分割方法依赖的特征是图像里同质区域的全局信息,提高了分割的准确性,并是分割方法有了一定的抗噪能力。本文首先介绍了基于主动轮廓模型的图像分割方法的产生和发展背景,然后详细阐述了水平集的基本理论和基于Mumford-Shah模型的图像分割方法。最后针对扫描文档图像的特点,依据分割结果的需要,本文采用基于简化Mumford-Shah模型的C-V图像分割方法进行文本行和自然段落的轮廓提取,并分析了本文算法的优缺点。文章中详细介绍了算法流程。因为在传统的水平集方法中,必须要进行复杂的水平集的重新初始化以确保水平集函数在迭代过程中保持为符号距离函数SDF,使得分割速度大幅减慢。为提高运算速度,实验中注意借鉴了李纯明的无需初始化的水平集方法,提高了图像分割的速度,并确保可得到一个有效的结果。本文引入李纯明的思路,经过数百次的实验证明该算法只需60秒,迭代10次以内就可以取得了较好的分割效果,而且是与语言的种类无关的的,适用于各种语言的文档图像。更多还原

【Abstract】 With the development of computer technology, the estrangement capability of computer has been enlarged many times, by kinds of input digital device, more and more documents are stored into computer and saved as bitmap form. There has been growing interest to the technology of convert these document images into a retrievable and editable form. For all these tasks, document image analysis comes in to being.Document segmentation is the major and basic technology in document processing and computer vision. Document segmentation got great academic and practical significance. The result of segmentation is better or not influence the following recognition and interpretation strongly. Therefore, many documentation segmentation methods have been developed and got successful in machine printed documents, processing of handwritten documents has still remains an open research field. Until now, the universal method to process all kinds of pictures has not being proposed. Most current documentation segmentation methods are based on that the document images are reasonably straight. Some segmentation approaches are depended on special language. Most proposes are sensitive to the topological changes of handwritten documents. Geometrical methods based on active contour model is not.Active contour model is a top-down processing with prior knowledge and provides a theoretically uniform frame work to a series of problems, such as contour extraction, stereo matching and object tracking. So the method has been successfully applied to image segmentation, medical image processing, human-computer interaction and many other research and practical fields. Level set methods which are based on Mumford-Shah model are excellent and important methods which are based on deformable model. Because of depending on global information of homogeneous regions in the image, they segment the images much more quickly and precisely.The paper introduces the background of active contour model. And then illuminates the foundation of the level set method and the image segmentation based on Mumford-Shah model. According to the characteristic of document, the author propose that the piecewise constant approximation of the Mumford-Shah model is very appropriate for the paragraph segmentation and text line segmentation. And the traditional level set methods must re-initialize level set functions costly so that level set functions can be closet to Signed Distance Function and image can be effectively segmented. But in order to be close to a Signed Distance Function, the time step must be small, and the evolution procedure is slowed down. The thesis introducing the Chunming Li’method of level set without re-initialization into them. The experiments indicate that the typically edges of our sample image will be picked up only no more than 10 iterations by using the proposed method. The segmentation tests for kinds of handwritten documents proved that the proposed method is very quick and universal.更多还原