基于Android平台文档文字识别技术的研究与实现

CORC > 自动化研究所 > 中国科学院自动化研究所 > 毕业生 > 硕士学位论文

题名	基于Android平台文档文字识别技术的研究与实现
作者	全远航
学位类别	工程硕士
答辩日期	2016-05-27
授予单位	中国科学院大学
授予地点	北京
导师	杜清秀
关键词	文档提取，文档图像预处理，文字识别，Android
中文摘要	近年来,伴随着移动互联网的快速发展以及移动端用户数量的猛增，越来越多的人们开始关注和尝试在移动设备上进行文档的版面分析与理解等相关的技术和应用，但由于受到移动设备的操作系统不统一、设计模式花样繁多、硬件性能参差不齐等方面因素的制约，文字识别技术并没有在移动设备上得到很好的应用和推广，因此，针对移动平台开展文档文字识别技术的研究与应用，具有非常重要的理论意义和实际应用价值。本文基于移动平台开展文档文字识别技术的研究与实现。在广泛了解并分析了国内外在移动端文档文字识别技术方面的研究现状及优缺点的基础上，针对移动端文档文字识别过程中所涉及到的关键技术点开展研究工作。在此基础上，融合多种关键技术的研发成果，实现了基于Android平台的文档文字识别系统，该系统能够从移动设备拍摄的照片中自动提取出文字内容，并将文字内容转化为电子设备可以阅读和使用的格式。本文的主要工作如下： 1.在文档图像预处理方面，针对移动设备拍摄的图片进行文档提取，实现了一种图片文档提取与畸变校正算法。针对光照不均等低质量图片，提出了一种改进的基于曲波变换的图片去噪和二值化算法。在文档文字分割方面，针对中英文混排的文档图片，提出了一种基于识别的中英文混排文档图片单字分割算法。 2.在文字识别技术方面，基于传统文字识别模型进行文字识别技术的研究，通过对单个字符图片进行归一化、特征提取、降维、分类器训练等步骤来达到文字识别的功能；同时，针对印刷体训练样本不足的问题，提出了一种快速生成印刷体样本的算法。 3.开发了基于Android平台的文档文字识别应用客户端。该客户端应用了本文提出的文档图像预处理技术、文字识别技术等关键技术的研发成果，实现了对手机拍摄的文档图片进行文档提取和文字识别的功能。
英文摘要	Character recognition technology has been one of the most important research areas in the field of pattern recognition. With the development of the mobile internet, the number of mobiles and users is increasing faster and faster. But the technology has not been widely used in the mobile due to the bad hardware performance. Whether the document image processing, or text recognition algorithm may not be able to get satisfactory results. So, the document character recognition technology still has a lot of problems worthy of study. This thesis analyzes the status of current character recognition technique in the world. And the key technologies in mobile character recognition process are researched. Finally, this paper realizes a text document recognition system on Android platform. It can extract text from photographs taken by mobile, and convert them into the format that electronic devices can use. The main contributions of this thesis are summarized as follows: 1. Research and improve the document image preprocessing technology Image preprocessing has a direct impact on the recognition performance. So this article firstly improves the document capture algorithm on mobile and puts forward to an algorithm for distortion picture taken by mobile. Secondly, document images may have complex background in form of non-uniform illumination or may be corrupted by noise. this thesis proposes an improved algorithm based on Curvelet transform for image denoising and image binarization. What’s more, this thesis put forward to an algorithm for character segmentation mixed Chinese and English. 2. Research the character recognition technology This paper takes the traditional machine learning methods，This method based on the traditional character recognition model, including the segmentation of characters, image normalization, feature extraction, dimension reduction, classifier training. In addition, in order to identify many types of printing fronts, this thesis also presents a fast algorithm for the generation of printed samples. 3. Develop the character recognition App On the Android Platform This thesis develops the OCR App based on Android platform. The App integrates document preprocessing technology, character recognition technology and realizes the function of extracting and recogniting characters in the document images taken by mobile phone. Experiments show that the App can provide relevant services for users, and achieve satisfactory results.
内容类型	学位论文
源URL	[http://ir.ia.ac.cn/handle/173211/11748]
专题	毕业生_硕士学位论文
作者单位	中科院自动化研究所
推荐引用方式 GB/T 7714	全远航. 基于Android平台文档文字识别技术的研究与实现[D]. 北京. 中国科学院大学. 2016.