文章导读
总览 评价 张斌 , 张洪刚 * , 李思远 , 谢廷 ( 信息与通信工程学院,北京邮电大学,100876; ) 摘要: 视频中字幕文字的自动定位及识别已经成为视频标注及检索系统中重要的组成部分。本文提出了一个自动定位、识别视频中字幕文字的系统,它主要由字幕文
张斌, 张洪刚*, 李思远, 谢廷
(
信息与通信工程学院,北京邮电大学,100876; )
摘要:
视频中字幕文字的自动定位及识别已经成为视频标注及检索系统中重要的组成部分。本文提出了一个自动定位、识别视频中字幕文字的系统,它主要由字幕文字定位、字符文字识别两个子系统构成。在字幕定位模块中,一种新颖的利用局部图像最大、最小值及其对比度信息的二值化方法被应用于系统中。大量实验证明该算法在增强文字前景、抑制复杂背景有良好的效果。得到文字前景图后利用多帧融合进一步滤除背景干扰,通过文字水平、竖直投影形成的波峰定位字幕位置。通过提取字符的局部边缘统计信息形成文字特征向量,计算向量的相似度进行字符识别;利用极大后验概率原理,我们对识别的关键词做进一步的后处理。在大量的实验基础上,本文取得了相对满意的结果,单字识别率接近0.9。
关键词:
模式识别与智能系统;字幕定位;LMM;投影;OCR
Zhang Bin, Zhang Honggang*, Li Siyuan, Guo Jie
(
Beijing University of Posts and Telecommunications, Beijing, 100876; )
Abstract:
Automatic caption location and recognition in videos is now recognized as a key component in the development of a advanced video annotation and retrieval systems. In this paper,a automatic caption location and recognition system is proposed,which is mainly composed of a caption location and character recognition subsystem.In the caption location module,a novel binarization technique named LMM,which makes use of the image contrast that is defined by the local image maximum and minimum is applied.Abundant experiment prove that this algorithm has good performance in suppressing the complex background in image. Multi-frames fusion is used to get a more accurately binarization image and projection method is adopted to segment the character later.Finally,extracting the characters's edge statistical information to form a feature vector, calculating the similarity between vectors to recognize the character. Do a post-processing to further improve the recognition rate. We achieved relatively satisfactory experiment results,our recognition rate is nearly 0.9 in various video .
Tag:
点此返回栏目查看更多>>>参考论文