基于过渡像素的视频图像文本检测与定位.pdfVIP

Vol. 38.No. 6 第 38 卷第6 期 湖南大学学报〈自然科学版〉 201 1 年 6 月 Jou口llllof Hunan University(Natural Sciences} Jun. 201 1 文章编号:1674-2974(2011)06-0069-06 基于过渡像素的视频图像文本检测与定位警 杨高波1f ，吴 潇1 ，张兆扬2 ，朱宁波1 (1.湖南大学信息科学与工程学院，湖南长沙 410082 , 2. 上海大学新型显示技术及应用集成教育部重点实验室，上海 200072) 摘 要:提出一种基于过渡像章的视频流人工文本检测与定位算法.该算法在水平和垂 直方向上提取过渡像素点，生成过渡固.通过块滤波器抑制背景的过渡像素，采用基于密度 的快速区域生长算法形成候选文本区域s 再利用改进的局部二进制模型(LBP)验证候选文 本区.仿真实验结果表明，本文的改进算法相对于 Kim 算法，字符定位的准确率是高. 关键词:文本处理s 视频流s 视频文本栓沸!，过渡囤 s 区域生长s 局部二进制模型 中国分类号:TP391 文献标识码:A A Transition Pixels Based Text Detection and Localization for Video Images YANG Gao-bolt I WU Xiao 1I ZHANG Zhao-yang2 I ZHU Ning-bo 1 (1. CoI1喀e of Infonnation Science 部td El咆h倒讪g， Hunan Univ , Changsha , Hunan 41∞82 ，。也lll; 2. Key Lab of Advanced Display and System Applica阳1S，.Minis町 of 剧ucation， Shanghai Univ , shanghai 2∞072 ， Clúna) Abstract: A transition pixel based detection and localization algorithm was proposed for the artificial texts in the video frame. The transition map was generated by extracting transition pixels in both horizon- tal and vertical directions. The transition pixels in the background were suppressed by block filtering , and candidate text regions were obtained by intensity based region growing. Finally , the candidate text regions were verified by improved local binary pattern (LBP). Experiment results have shown that , compared with Kims work , the proposed approach can achieve more accurate text detection and localization. Key words: text processing; video streaming ,video character detection ,transition map; region grow