【作者】 佟树成；

【导师】 王强；

【作者基本信息】 广西师范大学 ， 计算机软件与理论， 2007， 硕士

【摘要】 随着Internet的迅猛发展,利用Internet发布和交换信息更加频繁,同时数字图书馆和远程教育也成为热门领域,因此,信息资源的电子化是至关重要的一个问题。现有的OCR( Optical Character Recognition )技术虽然对中英文字和数字等都能获得很好的识别效果,但无法处理数学表达式。原因在于数学表达式具有复杂的二维嵌套结构,并且符号含义多样性,导致数学表达式在分割和结构分析等方面存在很多困难。为了方便读者和提高文献的利用率,设计一个系统,实现了将图像格式的印刷体数学表达式转换成可编辑的LaTeX格式的数学表达式。所完成的主要工作包括:1、设计实现了一个功能,快速方便地选取文献资料中要识别的数学表达式图像。2、图像预处理的功能是去除噪声,使反映符号本质特征的部分得到保留甚至突出出来,从而正确识别图像内容。3、特征提取与选择是图像识别的一个关键问题。本文对数学符号提取统计特征和结构特征组成45维特征集,对这样的特征集分类识别不受字体限制,而像素特征集受字体影响比较大,不同字体的同一个符号所提取的像素特征集差异可能比较大。并且设计实现K-L正交变换对特征集进行优化处理,去掉冗余信息,经过综合对比,最后确定39维特征集能取得比较理想的识别效果。4、支持向量机(SVM)是在统计学习理论的基础上发展起来的一种最新的机器学习方法。数学表达式识别是有限样本的多类识别问题,本文应用二类组合分类方法中的成对分类法(即一对一)进行符号识别。经过大量的实验,找到比较理想的参数,并与其它相关文献中的实验结果进行了比较,识别正确率有所提高。5、结构分析是数学表达式识别系统与字符识别系统的最大区别,也是难点之一。由于图像预处理时对符号进行了过分割,所以,设计一些规则检测合并多元结构的符号或函数型符号。然后,应用基准线的结构分析算法,使用树的存储形式,采用从上至下的分析策略,思想简单,容易实现,且识别正确率较高,速度较快,适合于数学表达式识别的实时系统。6、TeX系统是国际公认的最好的数学公式排版系统,LaTeX建立在TeX之上。因此,本文把图像格式的数学表达式转化成LaTeX格式的数学表达式。7、本系统全部功能完全在VC++6.0下实现,速度快,界面简洁,使用方便。更多还原

【Abstract】 With the development of Internet, using Internet to disseminate and exchange information is much more frequently. Digital library and distance learning through Internet are becoming hot research areas.So it is a crucial problem to make information resource electronic. The existing technology of OCR has a satisfying recognition effect for Chinese and English characters as well as digital symbols, but it is not efficient very much in recognition of mathematical expressions. Because of two-dimensional features of mathematical expressions and the variety of meanings of expressions, it has great difficulty in segmenting and structural analyzing of mathematical expressions. In order to convenience the readers and improve the using efficiency of literature, the system in this article could convert and reconstruct mathematics expressions in scanning the files directly into particular publication format,such as LaTeX.The major work has been completed as follows:1、A function is designed to realize the function of selecting the image of mathematical expression conveniently and rapidly.2、The function of image preprocessing is to remove noise, which can reflect the nature portion of the symbols prominently and then correctly identify image content.3、Image recognition is a key problem to feature extracting and selecting of image recognition.Based on the characteristics of mathematical symbols, statistics features and structural features are extracted from the mathematical symbols features to compose set of 45 dimensions,which has no restrictions of fonts, and great impact of fonts has been on pixel feature set.And the feature set is optimized by K-L orthogonal transformation in order to eliminate redundant information. At last, dimension 39 feature set are determined, which can achieve more satisfactory recognition results, by practice and comprehensive comparison.4、Support Vector Machine(SVM)is the latest machine learning methods which is developing on the basis of the statistical learning theory. Symbolic recognition of a mathematical expression is a limited sample of the various types problem.In this paper, symbols recognition was adopted the method of two types combinations classification which namely one against one. By a lot of experiments,ideal parameters are found and the experimental results are compared with other relevant documents,its correct rate increases than that of before.5、Structural analysis is the biggest difference between mathematical expression recognition system and characters recognition system, and it is also one of the difficulties in mathematical expression. Because of the over-segmentation made in a symbol of image segmentation in pretreatment, some rules are set up to detect and merge multiple structure characters or function characters.Structural analysis based on baseline algorithm is used, and the storage form of the tree and top-down analysis strategy are used, and all this good qualities are simple thinking, easy implement,high recognition accuracy and high speed,moreover it suit real-time identification of mathematical expression system.6、TeX system is recognized as the best mathematical formul typesetting system internationally,and LaTeX bases on TeX. Therefore, the mathematical expression of image is transformed into LaTeX form in the paper.7、The whole functional of system is actualized under the environment of Visual C++ 6.0.It can hold rapid speed, simple and convenient interface.更多还原