微博图像文字内容识别与感知

发布时间：2018-03-07 16:52

本文选题：微博图像　切入点：文字识别　出处：《北京交通大学》2014年硕士论文　论文类型：学位论文

【摘要】：近年来,微博网络的兴起与发展对科研领域带来了许多新的研究内容,并推动了自然语言处理、复杂网络社区发现等多个领域的研究。将多媒体中的内容提取并作为相关研究的样本,对以微博为研究对象的科研工作将具有良好的帮助作用。同时,对图片中的文字进行提取和识别也将极大地方便计算机自动化处理,在商业运用上也具有广阔的前景。图像文字识别主要作为模式识别中的一个重要分支,经历多年的发展后技术已经非常成熟,研究领域也逐渐扩展到自然场景文本识别、手写字符文本识别等方面,并获得了良好的应用。在文字识别的理论基础上,本论文主要工作为对文字识别中的文本定位和特征提取工作进行了研究和改进。并将图像文字识别技术应用到微博内容研究中,将提取和识别的文字保存,可以提供给其他研究领域进行进一步分析处理。本文在文本定位过程中结合图像的纹理和边缘特征,提出使用多尺度Gabor滤波器组对原图像进行变换处理,结合文本区域的先验知识对变换结果中的非文本对象进行过滤处理。边缘提取步骤中,使用Sobel算法对图像进行处理。将所获得的特征图像和边缘图像进行融合,并采用图像形态学的方法对融合结果作进一步的细节处理,从而获得文本区域。该方法提高了文本定位过程的准确性,在实际应用过程中的参数设置对经验值的依赖程度较小,具有较强的适应性。在单字符的特征提取中,采用了多尺度的Gabor滤波器组提取图像特征,形成一组包含多尺度多方向纹理特征的向量,最后使用SVM对所形成的纹理特征进行分类。最后,将上述方法应用到微博网络中,建立一个微博图像的采集和文字内容识别系统,完成微博图像的采集和逐步识别的功能,对提出的定位方法进行了验证。论文的工作得到了国家自然科学基金(No.61172072,61271308)、北京市自然科学基金(No.4112045)、高等教育博士点基金(No.W11C100030)、北京科技计划(No.Z121100000312024)和北京市教育委员会学科建设与研究生建设项目等课题的支持。
[Abstract]:In recent years, the rise and development of micro-blog network brings many new research content of the field of scientific research, and to promote the Natural Language Processing research complex network community detection and other fields. The content extraction in multimedia and as a research sample, will be helpful to micro-blog as the research object of the research work at the same time, the text in the image extraction and recognition will also greatly facilitate the computer automatic processing, but also has broad prospects in commercial use.
Image character recognition as an important branch of pattern recognition in the development, after many years of research in the field of technology has been very mature, gradually extended to the natural scene text recognition, handwritten character recognition and other aspects of the text, and has a good application in text recognition. Based on the theory, the main work of this thesis is on the text the positioning and feature extraction of character recognition has been studied and improved. And the image character recognition technology applied to the micro-blog research content in the extraction and recognition of text preservation, can be provided to other research fields for further analysis.
Based on the characteristics of the image edge and texture in the text positioning process, put forward the use of multi-scale Gabor filter of the original image is processed, non text objects on the text area transform combined with prior knowledge in filtering processing. The edge extraction step, make the image fusion image Sobel algorithm. And the edge image are obtained, and the details for further treatment of fusion results using the method of image morphology, so as to obtain the text region. This method improves the accuracy of text localization process, the parameters in the practical application in the process of setting the value of experience is less dependent and has strong adaptability.
In single character feature extraction, multi-scale Gabor filter banks are used to extract image features, and form a set of vectors containing multi-scale and multi-directional texture features. Finally, SVM is used to classify the textural features.
Finally, the above method is applied to the micro-blog network, and a micro-blog image acquisition and text content recognition system is established to complete the function of micro-blog image acquisition and gradual recognition, and verify the proposed location method.
This work is supported by the National Natural Science Fund (No.6117207261271308), Beijing Natural Science Foundation (No.4112045), Doctoral Fund of Higher Education (No.W11C100030), Beijing science and technology program (No.Z121100000312024) and Beijing Municipal Commission of education discipline construction and graduate construction projects to support issues.

【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.092;TP391.43

【参考文献】