手机拍照下题目分类算法的研究
发布时间：2018-11-14 20:13
【摘要】：随着科技的快速发展和网络资源的不断丰富,电脑和手机端的教育在日常生活中越来越发挥着举足轻重的作用。目前许多在线教育软件可以提供题目搜索功能,用户使用手机对题目进行拍摄,系统对拍摄的图像中的文字信息进行识别与检索,找到题库中与拍摄内容最接近的题目,并将答案与解答步骤反馈给用户,或提供相似题训练等服务。可见如果能够正确识别出文字对于提高用户体验有着重要意义,而拍照下的题目照片的文本区域识别和题目类别识别的题目分类算法是图片字符识别至关重要的一步,提高题目分类算法的准确率对于进一步的题目文本识别具有深远的意义。为了实现这一目的,本文针对文本区域定位算法、分类算法、系统整体设计和相关应用进行了研究和实践,具体的工作如下:1.针对文本区域定位算法的设计和实现,本文采用了笔画宽度变化的方法对手机拍照得到的图像中的题目所在的文本区域进行定位,使得后续操作只针对文本区域,减少了分割与识别的工作量,提高了准确度。2.针对定位好的文本区域分类算法,本文使用提取不止一个特征的方法,对已经定位好的区域特征进行全面的提取。然后使用从二分类支持向量机转化而来的三分类支持向量机将文本区域分类为数学,语文,英语三种类型。3.设计和实现了题目字符识别系统,鉴于光学字符识别有比较广阔的应用前景,与此同时为了验证前面两种算法的有效性,设计实现了基于图像字符识别的手机拍照下题目文本识别和题目数据库检索的系统。本文实现的两个算法,文本区域定位算法的召回率为79.04%,笔画准确度为79.59%,像素准确度为90.39%。准确率和运算速度优于其他文本定位算法。定位好的文本区域分类算法可以将分类的平均准确率达到92.32%,比传统分类器性能优越。
[Abstract]:With the rapid development of science and technology and the continuous enrichment of network resources, the education of computer and mobile phone plays a more and more important role in daily life. At present, many online educational software can provide the function of subject search, users use their mobile phones to shoot the title, the system recognizes and retrieves the text information in the captured image, and finds the title closest to the shooting content in the question bank. Feedback the answer and solution steps to the user, or provide similar problem training and other services. It can be seen that if we can correctly recognize the text, it is very important to improve the user experience, and the text area recognition and the topic classification algorithm of the subject photo are the most important steps in the image character recognition. It is of great significance to improve the accuracy of topic classification algorithm for further topic text recognition. In order to achieve this goal, the text region location algorithm, classification algorithm, the overall design of the system and related applications are studied and put into practice. The specific work is as follows: 1. Aiming at the design and implementation of text region localization algorithm, this paper uses the method of stroke width change to locate the text area of the topic in the image taken by mobile phone, so that the follow-up operation is only for the text area. Reduce the workload of segmentation and recognition, improve the accuracy. 2. Aiming at the text region classification algorithm, this paper uses the method of extracting more than one feature to extract the region feature. Then the text region is classified into three types: mathematics, Chinese and English by using the three-classification support vector machine which is transformed from the two-classification support vector machine. The title character recognition system is designed and implemented. In view of the wide application prospect of optical character recognition, in order to verify the validity of the two previous algorithms, The system of subject text recognition and subject database retrieval based on image character recognition is designed and implemented. The recall rate of text region location algorithm is 79.04, the stroke accuracy is 79.59 and the pixel accuracy is 90.39. The accuracy and operation speed are better than other text location algorithms. The text region classification algorithm with good location can achieve the average accuracy of 92.32, which is superior to the traditional classifier.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41
[Abstract]:With the rapid development of science and technology and the continuous enrichment of network resources, the education of computer and mobile phone plays a more and more important role in daily life. At present, many online educational software can provide the function of subject search, users use their mobile phones to shoot the title, the system recognizes and retrieves the text information in the captured image, and finds the title closest to the shooting content in the question bank. Feedback the answer and solution steps to the user, or provide similar problem training and other services. It can be seen that if we can correctly recognize the text, it is very important to improve the user experience, and the text area recognition and the topic classification algorithm of the subject photo are the most important steps in the image character recognition. It is of great significance to improve the accuracy of topic classification algorithm for further topic text recognition. In order to achieve this goal, the text region location algorithm, classification algorithm, the overall design of the system and related applications are studied and put into practice. The specific work is as follows: 1. Aiming at the design and implementation of text region localization algorithm, this paper uses the method of stroke width change to locate the text area of the topic in the image taken by mobile phone, so that the follow-up operation is only for the text area. Reduce the workload of segmentation and recognition, improve the accuracy. 2. Aiming at the text region classification algorithm, this paper uses the method of extracting more than one feature to extract the region feature. Then the text region is classified into three types: mathematics, Chinese and English by using the three-classification support vector machine which is transformed from the two-classification support vector machine. The title character recognition system is designed and implemented. In view of the wide application prospect of optical character recognition, in order to verify the validity of the two previous algorithms, The system of subject text recognition and subject database retrieval based on image character recognition is designed and implemented. The recall rate of text region location algorithm is 79.04, the stroke accuracy is 79.59 and the pixel accuracy is 90.39. The accuracy and operation speed are better than other text location algorithms. The text region classification algorithm with good location can achieve the average accuracy of 92.32, which is superior to the traditional classifier.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41
【相似文献】
相关期刊论文 前10条
1 景宁,刘雨,彭甫阳;一种实用外分类算法—快速分类-折半插入算法的研究及实现[J];小型微型计算机系统;1988年09期
2 郑智捷;幻序合并分类算法[J];计算机学报;1984年05期
3 刘t，
本文编号：2332182
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2332182.html
