字形轮廓的精确表示及其在文字向量化中的应用  

Exact Representation of Glyph Contours and its Application in Text Vectorization

作　　者：何涛[1] 冯洁[1,2] 周秉锋[1,2] 

机构地区：[1]北京大学计算机科学技术研究所,北京100871 [2]北京大学机器感知与智能教育部重点实验室,北京100871

出　　处：《影像技术》2011年第3期19-23,共5页Image Technology

基　　金：国家自然科学基金项目NO.6097354资助

摘　　要：从二进制图像中提取轮廓是许多图像处理应用中的重要内容,例如:扫描文档图像的向量化、对象分割、模式识别、手写文档和AutoCAD绘图中的内容解释、手绘卡通动画等。本文提出的轮廓跟踪算法能够生成具有以下特性的精确轮廓描述:首先,它能够有效地处理单像素宽度以及自相交的字形轮廓;其次,在轮廓跟踪的过程中,该算法可以将内、外轮廓与相应的对象区域相关联(例如字母‘B’由一个外轮廓与两个内轮廓组成),这将有利于对象的后续处理过程;第三,与传统的8-邻域或4-邻域连通轮廓跟踪算法相比,本文的方法能够精确地捕捉到字符的形状和尺度信息,这对于提高对象的保真度是十分重要的;最后,本文的算法简明,易于实现。Contour extraction from binary image is important for many image processing applications,e.g.scanned document image vectorization,object segmentation,pattern recognition,automatic content interpretation of handwritten documents and AutoCAD drawings,hand-drawn cartoon animation etc.The proposed contour tracing algorithm generates exact representation for contours,which has the following features： Firstly,it can effectively deal with one-pixel width glyphs,as well as self-intersection glyphs;Secondly,during contour tracing the algorithm can associate the outer and inner（i.e.the letter ‘B’ has one outer and two inner contours） contours with the corresponding object regions,which favors subsequent manipulation of the object;Thirdly,compared with traditional 8 or 4-neighbour connected contour tracing,our methods can exactly capture the glyph shape and size information,which is important in preserving the fidelity of objects;Lastly,the algorithm is concise and easy to implement.

关 键 词：文档图像 向量化 字形 轮廓 眼腺 

分 类 号：TP317.4[自动化与计算机技术—计算机软件与理论]