自然场景下基于连通域检测的文字识别算法研究  

Research on Character Recognition Algorithm Based on Connected Domain Detection in Natural Scene

作　　者：刘新瀚[1] 钱侃[1] 王宇飞[1] 朱向霄 孙知信[1] 

机构地区：[1]南京邮电大学物联网学院,江苏南京210003

出　　处：《计算机技术与发展》2015年第5期41-45,共5页Computer Technology and Development

基　　金：国家自然科学基金资助项目(60973140;61170276;61373135);江苏省产学研项目(BY2013011);江苏省科技型企业创新基金项目(BC2013027);江苏省高校自然科学研究重大项目(12KJA520003)

摘　　要：自然场景中由于文字背景的复杂性等原因,给文字识别工作带来了极大的困难。文中提出一种边缘检测与连通域分析相结合的算法以识别自然场景中的文字,提高文字识别的准确率与召回率。首先采用ColorRoberts算子直接对彩色图像进行边缘检测,从而避免彩色图像转换为灰度图像过程中的信息丢失现象;然后对检测出的图像边缘进行去除长直线、去除孤立的噪声点、形态学运算的后续处理操作;最后经过连通域标记、分析,提取出文本区域。通过仿真实验,结果表明了该算法的合理性和有效性。Due to the complexity of text background in natural scene,it brings great difficulties to the character recognition. In this paper, propose an algorithm combined edge detection and connected component analysis to identify the text in natural scene,which improves the accuracy and recall rate of text recognition. Firstly,the ColorRoberts operator is used to detect the edge of color image directly in this al-gorithm,thus avoiding the information loss in the process of conversion from a color image to a gray image. Secondly,the subsequent pro-cessing operation,such as removing the long straight lines,removing isolated noise points and morphological processing,is actualized on the image edge which has been detected. Finally,through the connected domain labeling and analyzing,the text regions are extracted. Simulation results are given,which shows that the algorithm is reasonable and effective.

关 键 词：边缘检测 ColorRoberts算子 连通域标记 连通域分析 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]