CASIA OpenIR  > 毕业生  > 硕士学位论文
文本定位研究
Alternative TitleResearch of Text Location
欧文武
Subtype工学硕士
Thesis Advisor刘昌平
2004-07-01
Degree Grantor中国科学院研究生院
Place of Conferral中国科学院自动化研究所
Degree Discipline模式识别与智能系统
Keyword文本定位 自然场景 Gabor滤波器 版面分析 图像处理 Text Location Scene Image Gabor Filter Layout Analysis Image Processing
Abstract通过OCR(0ptical Character Recogrlition) 于信息检索、智能交通和票据、证件处理等方面有重要意义。通常在OCR系统 识别图像上的文字时,要先找出文字区域位置,我们把从图像上找出文字区域 的过程叫文本定位,文本定位目前依然是没有解决好的难题。本文根据自然场 景图像、视频图像(帧)和手写信封图像的文本特点分为场景文本定位、视频文 本定位和手写信封版面分析分别做了研究。论文的主要工作包括以下几个方面: 提出了一种自然场景文本定位方法。首先利用文字边缘密度和形态学运算 找出备选文本区域,然后通过颜色空间可分、连通域分布和投影分析等后处理 方法抑制背景,提高定位精度。试验表明,本文提出的方法能够在多种自然场景 图像上比较准确地找出文本位置。 将Gabor滤波器引入视频文本定位,并提出了一种非常实用的Gabor特征 选择方法。在选取Gabor特征时,本文先通过Fisher准则选择Gabor滤波器参 数,挑出备选Gabor特征,然后通过BP网络的分类结果找出分类结果最好的 特征组合,即Gabor特征向量。试验结果表明通过上述方法找出的Gabor特征 向量能够非常有效地区分文本和非文本区域。 提出了一种非常实用的手写信封版面分析方法。先去除信封图像上的图案 和邮票等冗余信息,降低信封版面复杂程度,然后通过递归投影和连通域分析 结合的版面分析算法找出信封图像上的文字区域,文中还提到通过拒识的方法 剔除部分图像质量太差的信封。对上万个信封图像的测试表明,本文提出的算 法能够正确处理大部分各种版面的信封图像。
Other AbstractRecognizing text in image by OCR system is very important for information retrieval, intelligent communication, and note or certificate procession. Normally, before text recognized by OCR system, we have to locate the text in image, and the process of finding text in image is called as Text Location, which is still a hard problem. Based on the feature of scene images, video images (frames) and envelope images, we made research on Text Location in Scene Images, Text Location in Video Frames and Layout Analysis of Envelope Images respectively. The main work of this paper is as following: We propose a novel method of text location in scene images. Locate the candidate text region by edge intensity and morphological operation, then several post-processing algorithms, such as separability of color space, distribution of connected components and projection profile, is used to exclude background and refine location result. The experiment proves that our method can exclude most background region, and can locate text regions in image accurately. In this paper, we introduce the Gabor filters to text location in video frames, and propose a practical way on feature selection of Gabor filters. By Fisher criterion, we select the Gabor filters' parameters firstly, and then pick out a set of Gabor filters, which have the best classification result by BP Network. And the experiment proves that the Gabor eigenvector, selected by our method, can distinguish text and non-text region efficiently. We propose a practical method for layout analysis of hand-written envelope images. By eliminating the pattern and stamp in envelope images, we decrease the complexity of envelope images' layout greatly, and then by the layout analysis algorithm, which combines the projection method and connected component analysis, we find out the text region in envelope images. And we also propose the way to exclude some envelope images of bad quality. By the test of thousands of envelope images, our algorithm can deal with most envelope images properly.
shelfnumXWLW783
Other Identifier783
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/6778
Collection毕业生_硕士学位论文
Recommended Citation
GB/T 7714
欧文武. 文本定位研究[D]. 中国科学院自动化研究所. 中国科学院研究生院,2004.
Files in This Item:
There are no files associated with this item.
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[欧文武]'s Articles
Baidu academic
Similar articles in Baidu academic
[欧文武]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[欧文武]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.