文本图像的几何畸变校正技术研究
- 1、下载文档前请自行甄别文档内容的完整性,平台不提供额外的编辑、内容补充、找答案等附加服务。
- 2、"仅部分预览"的文档,不可在线预览部分如存在完整性等问题,可反馈申请退款(可完整预览的文档不适用该条件!)。
- 3、如文档侵犯您的权益,请联系客服反馈,我们会尽快为您处理(人工客服工作时间:9:00-18:30)。
河北工业大学
硕士学位论文
文本图像的几何畸变校正技术研究
姓名:吴丽平
申请学位级别:硕士
专业:通信与信息系统
指导教师:于明
20091101
分类号:TP391密级:
U D C:编号:
学位论文
文本图像的几何畸变校正技术研究
吴丽平
指导教师姓名:于明教授河北工业大学
申请学位级别:硕士学科、专业名称:通信与信息系统论文提交日期:2009年11月论文答辩日期:2009年12月
学位授予单位:河北工业大学
答辩委员会主席:
评阅人:
2009年11月
Dissertation Submitted to
Hebei University of Technology
for
The Master Degree of
Communication and Information Systems
RESEARCH ON GEOMETRIC DISTORTION CORRECTION TECHNOLOGY FOR DOCUMENT IMAGE
by
WU Liping
Supervisor:Prof.YU Ming
November2009
文本图像的几何畸变校正技术研究
ii
文本图像的几何畸变校正技术研究
摘要
在使用扫描仪或数码相机获得文本图像时,由于文本表面倾斜、弯曲或人为操作时产生的拍摄视角的倾斜等原因,使所得到的文本图像存在几何畸变,这些畸变对文字处理软件如OCR识别、数字文档版面分析等的识别和分析工作带来极大的困难,可能会导致这些软件根本无法识别该文本图像。因此,必须对变形文本图像进行必要的校正。目前,已经有很多几何畸变校正的方法应用于畸变的文本图像。但是这些方法都是针对某一种或两种几何变形有效,而没有绝对通用的算法,需要对不同变形类型的图像采取不同的有效校正算法。
为实现几何畸变文本图像的自动校正和批量处理,论文在讨论文本图像二值化、去噪技术的基础上,重点对文本图像几何畸变的自动检测及分类的方法进行研究。提出了基于数学形态学理论与曲线拟合方法的自动检测和分类方法,实现对文本图像几何畸变的自动检测并对其畸变类型进行分类,为后续的几何畸变自动校正奠定了基础。然后论文分别针对倾斜变形、透视变形和扭曲变形三种典型畸变文本图像,在对现有的校正算法原理、性能及适用范围进行分析的基础上进行了研究,并提出了一些改进方法。
在Matlab环境中对文本图像几何畸变自动检测及分类算法进行了验证,试验结果表明该算法能有效检测和识别文本图像的几何畸变及其类型,检测识别率达到96%以上,尤其是对扭曲变形的文本图像识别率很高。且该算法实现简单,计算量小,并具有较强的鲁棒性。然后通过试验比较文本图像校正前后的OCR文字识别率,分别对三种变形文本图像几何校正算法的改进算法进行了验证。结果表明,与原有算法相比,改进算法所得校正图像的OCR识别率不低于原有算法,但所用时间降低,基本满足实时性要求。在数字化和信息化的现代,该文本图像几何畸变自动校正系统具有广阔的应用前景。
关键词:OCR,文本图像,几何畸变,倾斜变形,透视变形,扭曲变形,数学形态学
iii
文本图像的几何畸变校正技术研究
RESEARCH ON GEOMETRIC DISTORTION CORRECTION TECHNOLOGY FOR DOCUMENT IMAGE
ABSTRACT
When obtain the document images used a scanner or digital camera,there is likely to exist geometric distortions because of the bending of document or the tilt in shooting angle caused by man-made operations.These deformations create great difficulties for word-processing software, such as OCR recognition or digital document layout analysis,even may lead to the software do not work at all.So,the document images must be corrected before that.At present,there are already a lot of correction methods applied to geometric distortion document image.However, these methods are effective against only one or two kinds of geometric distortion,and there is no absolute general algorithm.Therefore,we need to take different approaches for different deformation.
To achieve the automatic document image geometric distortion correction and batch processing,this paper put an emphasis on the automatic detection and classification method to geometric distortion of document images,based on the technique of document image binarization and de-nosing.The algorithm combines the mathematical morphology processing and curve fitting methods,to detect the geometric distortion of document images and classify the types of their distortion automatically,which laid foundation for automatic correction of geometric distortion following.And then some exploratory research had been done on the correction algorithms to document images with geometric distortion,which includes inclination deformation,perspective distortion and torsion deformation,based on the analysis of the theory, performance and scope of application of existing bined the detection and classification method,we made some improvement in the correction algorithms.
Experiments had been done to the automatic detection and classification method of document images with geometric distortion in the Matlab environment.And it was verified by experimental results that the algorithm has higher recognition rate reached96%or more,especially for the torsion deformation.And the algorithm is simple,with small amount of computation,and has strong robustness.Then the three kinds of geometric distortion correction algorithms with improved were verified,through compared the OCR recognition rates before and after correction of the document image respectively.The results showed that the improved algorithm effectively reduce the run time to basically meet the real-time requirements,without declining the OCR recognition rates.In the digital and information-based modern,the automatically geometric distortion correction system to document image has broad application prospects.
KEY WORDS:OCR,Document image,Geometric distortion,Inclination deformation,
Perspective distortion,Torsion deformation,Mathematical morphology iv