扫描阅卷系统中模板定制和图像聚类方法研究.docVIP

扫描阅卷系统中模板定制和图像聚类方法研究 　　摘要：为了适应答题卡多样化需求和提高答题卡图像识别的准确率，提出了扫描阅卷系统中模板定制和图像聚类方法。首先基于人机交互方式进行模板定制，定义填涂区域属性和答题卡结构信息，并开发了模板制作器，实现答题卡模板文件的制作和管理；其次给出基于Kmeans改进算法的扫描阅卷系统中图像聚类方法，选择局部聚集密度最大的数??点作为初始聚类中心以得到全局较优的聚类结果，并通过计算区分度进行聚类结果评价；最后基于VC++和MS SQL Server2000开发了基于Kmeans改进算法的扫描阅卷系统，并对该系统进行了实验测试。测试结果表明，采用Kmeans改进算法进行扫描阅卷时能够得到稳定的图像聚类结果，大大提高了客观题阅卷准确率，具有较高的实用价值。 　　关键词：扫描阅卷系统，模板定制，Kmeans改进算法，图像聚类 　　中图分类号：TP391文献标识码：A 　　Abstract：To adapt to the requirement of exam card diversity and enhance the image recognition accuracy of exam card，template customization and image clustering method in the exam card scanningreading system has been proposed.At first，template customization based on human computer interaction has been given and the attribute of the filling zone and structural information of exam card has been defined.And template customization tool has been developed to realize template customization and management.Then image clustering method based on Kmeans improved algorithm has been given.To obtain clustering results of the global optimum，data points with the maximum local gathering density has been chosen as the initial clustering centers.And clustering results has been evaluated by calculating the item discrimination.At last with VC++ and MS SQL Server2000 exam card scanningreading system based on Kmeans improved algorithm has been developed and the system has been tested.It has been shown that that stable image clustering results can be obtained with the exam card scanningreading system based on Kmeans improved algorithm.And image recognition accuracy of exam card has been improved.The exam card scanningreading system has better utility value. 　　Key words：exam card scanningreading system；template customization；Kmeans improved optimization；image clustering 　　1引言 　　阅卷是考试中的一个重要环节，传统机器阅卷需要光电阅卷机和专用机读卡，虽然具有阅卷速度快和适合大规模考试的优点，但只能采用专用答题卡进行扫描阅卷，存在产品价格高和无法灵活定制试卷答题模板等缺陷[1-3]。另外，扫描阅卷系统对图像输入设备采集答题卡图像自动进行图像处理和识别，其操作简单且成本低廉，越来越得到广泛应用[4，5]；但在扫描阅卷系统中由于扫描得到的图像质量存在个体差异，填涂未擦除干净或填涂深浅不一会增加客观题的误判率。