基于SVM的流程图像角点分类  

Corner classification of flowchart based on SVM

作　　者：孙连山[1] 张沙沙 侯涛[1] 赵晓[1] SUN Lian-shan;ZHANG Sha-sha;HOU Tao;ZHAO Xiao(College of Electrical and Information Engineering,Shaanxi University of Science&Technology,Xi′an 710021,China)

机构地区：[1]陕西科技大学电气与信息工程学院,陕西西安710021

出　　处：《陕西科技大学学报》2018年第2期147-153,共7页Journal of Shaanxi University of Science & Technology

基　　金：国家自然科学基金项目(61601271);陕西省教育厅专项科研计划项目(17JK0087)

摘　　要：针对现有流程图像识别研究未能较好地处理文图粘连及断边问题,结合流程图像角点特征不受图文粘连和断边影响的性质,提出针对流程图像角点的自动检测和分类方法,为基于角点分类识别流程图结构并进一步开展流程图像理解研究奠定基础.首先,结合流程图像角点分析和流程图结构描述,定义了流程图像的角点分类模型;然后,根据连通域面积实现文图分割提取流程结构图像,综合运用经典算法检测流程结构图像的角点;最后,提取角点邻域的网格特征和外围特征、采用有监督的方式训练SVM分类器实现角点分类.针对CLEP-IP的公开流程图像集进行测试,实验证明角点分类准确率达到91.6%.Existed flowchart recognition works failed to deal with touched line-texts and broken lines.Corners in flowchart images are not affected by touched line-texts and broken lines and can be used to recognize logical structure of flowcharts to enable automatic comprehension of flowchart images.An approach was proposed to automatically detect and classify the corners in flowchart images.Firstly,typical corner types are identified and a corner-based structural semantic model of flowchart is defined through analyzing flowchart structures from the perspective of corners.Secondly,the structural layer of a flowchart image is extracted using the area of connected components and typical corner detectors are used synthetically to identify corners in flowchart images.Finally,grid features and peripheral features of the neighborhood of corners are extracted for training a SVM-based corner classifier in a supervised learning process.The corner classifier is validated using a public dataset from CLEP-IP.The overall corner recognition rate is 91.6%.

关 键 词：流程图 角点检测 特征提取 SVM 角点分类 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]