视频中滚动字幕的检测与跟踪  

Detection and Tracking of Rolling Caption in Video

作　　者：李钦瑞 吕学强[1] 李卓[1] 刘坤 

机构地区：[1]北京信息科技大学网络文化与数字传播北京市重点实验室,北京100101 [2]北京拓尔思信息技术股份有限公司,北京100101

出　　处：《小型微型计算机系统》2015年第3期631-636,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61171159;61271304)资助;北京市教委科技发展计划重点项目暨北京市自然科学基金B类重点项目(KZ201311232037)资助;北京市属高等学校创新团队建设与教师职业发展计划项目(IDHT20130519)资助;北京信息科技大学网络文化与数字传播北京市重点实验室开放课题(ICDD201303)资助

摘　　要：视频中滚动字幕通常表达一段完整信息,对其进行检测与跟踪是字幕识别与应用的重要前提.通过分析视频中滚动字幕的特征,在Sobel边缘检测的基础上,提出规则和统计结合的字幕检测方法及垂直边缘特征最小差平方和的字幕跟踪方法.从视频全局角度出发,对相邻帧的边缘图像进行差分,选取水平边缘特征并根据规则检测字幕区域的候选边界,通过统计边界出现频数,定位滚动字幕区域;计算相邻滚动字幕条在不同偏移值下垂直边缘特征差平方和,取其最小值作为相邻字幕的偏移量,从而实现字幕的跟踪.实验结果显示,该文方法从含有多种字幕类型的视频中检测出滚动字幕的准确率可达96.61%、字幕跟踪的准确率达99.43%,充分说明文中方法的有效性.The detection and tracking of rolling caption, which usually deliver complete messages in video, are the premise of caption recognition and application. Analyzing the features of rolling caption and Sobel edge images, this paper proposes a caption detection method based on the integration of rules and statistics and a caption tracking method using the vertical edge features. From the overall perspective of video,this paper differences the adjacent frame,extracts horizontal edge features to detect the candidate edge of the cap- tion＇s area according to certain rules,and counts the frequency of the edge＇s occurrence to locate the rolling caption＇s area. Then this paper calculates the sum of square of adjacent caption＇s vertical edge features under different offsets,and takes the minimum value as the offset of adjacent captions to realize the caption tracking. Experimental result shows that our method can achieve 96.61% precision in detecting rolling caption from video containing various kinds of caption,99.43% precision in tracking rolling caption.

关 键 词：滚动字幕 字幕检测 字幕跟踪 边缘特征 规则与统计 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]