一种面向问卷图像的版面分析算法

发布时间：2018-04-16 13:21

本文选题：行分割 + 连通区域　；参考：《软件学报》2017年02期

【摘要】：针对目前已有的问卷图像版面分析算法无法自动识别信息填写区域和无法处理无固定格式的问卷图像等问题,提出了一种连通区域和神经网络相结合的问卷图像版面分析算法.首先获得扫描得到的问卷图像的中心有效图形,接着提出并应用了一种针对问卷图像的快速倾斜矫正方法,对中心有效图像进行倾斜矫正;再利用水平投影进行行分割得到问卷行;然后提取每个问卷行的首个连通区域判断是否存在表格区域即表格问卷行,若存在表格问卷行,则对其进行表格区域分布分析和表格类型判断,得到可能的答案区域,否则直接对文本问卷行进行分析,得到可能的答案区域;最后利用神经网络判断筛选区域的类型,得到最终的答案填写区域.针对问卷图像的实验结果表明,该算法可以准确地识别各种问卷图像中的信息填写区域.
[Abstract]:Aiming at the problems that the existing algorithms of questionnaire image layout analysis can not automatically identify the information filling area and can not deal with the questionnaire images with no fixed format, this paper proposes a new algorithm of questionnaire image layout analysis based on the combination of connected region and neural network.Firstly, the central effective image of the questionnaire image is obtained, and then a fast skew correction method for the questionnaire image is proposed and applied to the center effective image.The first connected region of each row is extracted to determine whether there is a tabular area, that is, the questionnaire row, if there is a tabular questionnaire row.Then we analyze the distribution of tables and the types of tables, and get the possible answer areas, otherwise, we analyze the lines of the questionnaire directly and get the possible answer regions. Finally, we use the neural network to judge the types of the selected regions.Get the final answer and fill in the field.The experimental results on the questionnaire images show that the algorithm can accurately identify the information filling areas in various questionnaire images.
【作者单位】：西安交通大学软件学院;
【分类号】：TP391

【相似文献】