智能阅卷系统中图像处理相关技术研究

发布时间：2019-01-21 19:05

【摘要】：网上阅卷相比传统人工阅卷,结果更准确、阅卷效率更高、阅卷过程更安全,有极大的优势。网上阅卷系统在上世纪90年代被提出,经过几十年的发展,已普遍使用在大型考试当中。但中小型考试大多仍采用人工阅卷方式,在大型考试中也依然需要光标阅读机的辅助阅卷,成本非常高,后期维护困难,依赖题卡分离模式。而在题卡合一模式下则必须使用携带所有题目信息的复杂模板对试卷进行解析,占用大量资源,拉长了阅卷周期。另外,试卷图像的分割大多是定区域分割,不会根据考生实际答题区域进行智能分割。基于以上阅卷方式普遍存在的问题,本文提出了一种基于不同题卡模式的智能阅卷系统,对不同题卡模式分别采用不同算法对试卷进行倾斜校正,使用精简版模板进行定位,优化了试卷信息的智能识别,极大内提高了系统效率。主要包括以下几个方面:(1)提出了基于不同题卡模式的两种倾斜校正算法。一是针对大型考试中有同步头的答题卡提出了一种新的算法,准确得到了试卷图像倾斜角度;二是针对在中小型考试中没有任何同步头信息的答题卡,满足试卷形式的更多需求且提高了算法效率。(2)提出使用XML精简版模板进行粗定位。省去了用DOM解析XML文件的步骤,减轻了系统负担。(3)引入一维码携带考生信息并对一维码图像进行识别。传统考试考生需要手动填写个人信息效率低下,字符识别容易出错,本文引入一维条形码来记录考生信息并实现了一维码图像的定位识别,简化了识别考生信息的过程,提高了识别的效率和准确率。(4)优化了客观题的识别与分割算法。对客观题进行两次分割,并用OTSU法得到阈值作为参考,比较后得到考生选项,解决了之前的很多阅卷系统中客观题阅卷只分割不识别的缺陷。(5)提出一种新的主观题识别算法。在主观题阅卷中提出结合了竖直方向腐蚀运算的水平投影法,实现了更为智能的依据考生答题区域进行分割的算法。由实验和数据结果分析得出,本文算法有效的解决了阅卷系统中遗留的问题,系统的设计也更为灵活。可以根据试卷类型选择相应合适的倾斜校正算法并提升了算法性能。对试卷信息识别分割的相关算法进行了优化,进一步智能化了阅卷过程。
[Abstract]:Compared with the traditional manual marking, the result of online marking is more accurate, the marking efficiency is higher, the marking process is more secure and has great advantages. The online marking system was put forward in the 1990s. After decades of development, it has been widely used in large-scale examinations. However, most of the small and medium-sized examinations still use manual marking method, and in large scale examination, it still needs the help of the cursor reader. The cost is very high, the maintenance is difficult in the later period, and the separation mode of the item card is relied on. However, in the mode of integration of question and card, the paper must be analyzed by using the complex template with all the subject information, which takes up a lot of resources and prolongs the marking period. In addition, test paper image segmentation is mostly fixed area segmentation, not according to the actual answer area of candidates for intelligent segmentation. Based on the above problems, this paper presents an intelligent marking system based on different card patterns, which uses different algorithms to correct the test papers, and uses a simplified template to locate the test paper. The intelligent recognition of test paper information is optimized and the system efficiency is greatly improved. The main contents are as follows: (1) two skew correction algorithms based on different problem card modes are proposed. First, a new algorithm is proposed for the answer card with synchronous head in the large scale examination, which can get the angle of the image of the test paper accurately. The second is to meet more requirements of test paper form and improve the efficiency of algorithm for the answer card without any synchronous information in the small and medium-sized examination. (2) this paper proposes the use of XML template for rough positioning. The steps of parsing XML files with DOM are saved and the system burden is lightened. (3) One-dimensional code is introduced to carry the examinee information and the one-dimensional code image is recognized. Traditional examinees need to fill in personal information manually, and character recognition is easy to make mistakes. This paper introduces one-dimensional bar code to record examinee information and realize location recognition of one-dimensional code image, which simplifies the process of identifying candidate information. The efficiency and accuracy of recognition are improved. (4) the objective problem recognition and segmentation algorithm is optimized. The objective questions are divided twice, and the threshold value obtained by OTSU method is used as a reference. After comparison, the candidates' options are obtained. It resolves the defect that the objective questions in many previous marking systems are segmented and unrecognized. (5) A new subjective problem recognition algorithm is proposed. The horizontal projection method, which combines the vertical corrosion operation, is put forward in the subjective examination paper, which realizes a more intelligent algorithm of dividing the test questions according to the examinee's answer area. The results of experiments and data analysis show that the algorithm can effectively solve the problems left over in the marking system, and the design of the system is more flexible. An appropriate skew correction algorithm can be selected according to the test paper type and the performance of the algorithm is improved. This paper optimizes the algorithm of paper information recognition and segmentation, and further intelligentizes the marking process.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：G434;TP391.41

【相似文献】