自然场景下基于区域检测的文字识别算法  

Character Recognition Algorithm Based on Region Detection in Natural Scene

作　　者：陈梓洋 王宇飞[1] 钱侃[1] 张超[1] 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7期230-233,共4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：由于自然场景的图像具有复杂性和不确定性,导致自然场景下文字识别相比于文本识别更加复杂和困难,利用现有识别算法对自然场景图像进行识别,其识别效率较低,识别效果不理想。为提高识别率,文中首先利用分水岭算法对原图像进行初处理,再对预处理后的图像进行特征提取,划分区域,最后利用ISODATA算法进行第二次处理,最终对图像中的文字进行识别。对提出的算法进行仿真和实验,实验结果表明,采用该算法后,自然场景下文字的识别效果明显,识别率高。Because the natural scene text has the characteristic such as the uncertainty and complexity, the natural scene is becoming diffi- cult and important in character recognition. By using the existing recognition algorithm for identification of natural scene images, the rec- ognition rate is low and the recognition effect is not ideal. In order to improve the recognition rate,firstly use watershed algorithm to finish the fast original image processing, then extract feature and partition the image after preprocessing, finally utilize ISODATA algorithm to process second times and identify the text in the image. By simulating and experimenting on the proposed algorithm, the experimental re- suits show this algorithm can recognize the text in natural scene efficiently with the high recognition rate.

关 键 词：分水岭算法 ISODATA算法 基于区域的检测算法 自然场景 文字识别 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]