基于角点密集度的维吾尔视频文字区域定位  

Text regions localization of Uyghur videos based on corners intensity

作　　者：依再提古丽.克热木 地里木拉提.吐尔逊 艾斯卡尔.艾木都拉[1] EZZATGUL Keram;DILMURAT Tursun;ASKAR Hamdulla(College of Electronic Engineering, Guangxi Normal University, Guangxi Guilin 541004,China)

机构地区：[1]新疆大学信息科学与工程学院,新疆乌鲁木齐830046

出　　处：《电视技术》2017年第11期225-231,246,共8页Video Engineering

基　　金：国家自然科学基金项目(61461049)

摘　　要：针对如何定位视频序列中维吾尔文字区域,提出了Harris角点检测与角点密集度相结合的方法。通过Harris算法得到的角点分布图进行背景角点的滤除,再根据角点间距离确定切分点,得到角点图像的分层、分块区域,并对每一个区域块确定文字的基线位置,进行膨胀,得到候选文字区域,采用启发式规则和连通域合并获取最终文本。分别对单行和多行文字区域视频帧进行试验,得到查准率分别为92.26%和76.01%,共有2305个文字区域,平均正确检测率达到93.08%。测试结果表明,该方法解决了视频帧图像中低对比度、背景复杂、含有艺术字体、文字区域排布不规律等问题。To locate Uyghur text regions in video sequences,this paper proposed a new algorithm based on combining Harris comers and comers distance classification.Firstly,detect the Harris comers and filtered out the non text comers.Based on comers distance get the segmented points used stratifying comer regions,then determine the baseline of every stratified region,and carry on morphological expanding operations to the comers within the baseline zone get the candidate text area.Finally,used heuristic rules and merging the connected components get the final text regions.In this paper,experimental objects are single text area and multi line text area in video frames,The precision ratio was 92.26％ and 76.01％,there are 2305 text regions,average correct detection rate reached 93.08％.The experimental results show that proposed method solves the video frame images of low contrast,complex background,with art font,and irregular arrangement of the text regions.

关 键 词：维吾尔文字 视频图像 HARRIS角点 角点间距离 基线特征 连通域合并 

分 类 号：TN244[电子电信—物理电子学]