改进的视频帧中文本区域定位算法  

Improved algorithm for text region localization in video frames

作　　者：张海云[1] 张凤荔[1] 刘伟[1] 万明成[1] 

机构地区：[1]电子科技大学计算机科学与工程学院,成都611731

出　　处：《计算机工程与应用》2011年第7期189-192,共4页Computer Engineering and Applications

基　　金：国家高技术研究发展计划(863)No.2006AA01Z41~~

摘　　要：针对视频帧图像背景复杂、字体大小变化较大等特点,提出一种改进的视频帧中文本区域定位算法。算法设计并实现一个交叉点检测算法,利用交叉点密度等边缘特征去除大部分非文字边缘以降低背景边缘对文本区域的影响,并对剩余边缘进行膨胀以形成候选文本区域,结合文本区域特征和支持向量机算法区分文本区域和非文本区域。实验表明,该算法可以提取视频帧中90%以上的文本区域,并且文本区域定位准确率达92.0%。Contraposing the characters of images in video,such as complicated backdrop and big diversification of font,this paper proposes an improved algorithm for text region localization in video.The algorithm designs and implements a detecting cross points algorithm at first.Then the algorithm recognizes and deletes most of non-text edges with the features of edge like the density of cross points,in order to reduce the influence of backdrop on text region,and then expands the rest edges to form candidate text regions.The text region and non-text region are distinguished by combing the features of text region and a SVM classifier.The experiments show that the algorithm can extract more than 90% of text regions contained in video frames and the precision can reach up to 92.0%.

关 键 词：边缘检测 交叉点检测 图像膨胀 支持向量机 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]