基于主题字幕提取的新闻视频检索研究

【作者】 王艳；

【导师】 王建宇；

【作者基本信息】 南京理工大学 ， 系统工程， 2008， 硕士

【摘要】 近年来,基于内容的多媒体信息检索已经成为了一个热门的研究领域,新闻视频检索作为其中一部分也得到了广泛的研究。新闻视频是人们获取信息的主要媒体,但如何快速、准确地从海量的新闻视频数据中,找到所需的内容成为一个迫切需要解决的问题。本文以新闻视频为研究对象,完成了以下几个部分:镜头分段、关键帧提取、主题字幕提取、新闻故事分段和新闻视频检索,具体工作如下:(1)镜头分段是后续视频处理分析的前提,本文改进了基于双重比较的镜头分段算法,将自适应阈值运用于双重比较法,完成了对新闻视频的突变和渐变镜头检测,具有更高的准确性。由于主题字幕最大限度地反映了新闻的主要内容,本文提出了基于主题字幕帧的关键帧提取算法,提取的关键帧基本上可以代表镜头内容。(2)主题字幕提取包含三部分内容。首先根据新闻主题字幕的特点,设计了基于3/10时空切片的字幕帧检测算法,取得了较好的检测效果;然后采用基于小波变换和支持向量机的字幕区提取算法,实现了主题字幕的准确定位;最后对主题字幕进行插值放大、二值化、字符分割及OCR文字识别,识别效果较好。(3)针对新闻结构的特点,设计了基于主题字幕提取的新闻故事分段算法,在主题字幕提取的基础上,通过静音检测、主持人镜头检测,把连续的新闻视频分割成一个个的新闻故事,能够达到令人满意的效果;根据新闻分段结果,以及主题字幕的标注,实现了基于关键字的新闻视频检索。更多还原

【Abstract】 Content-based multimedia retrieval has been a hot research field recent years. As a part, news video retrieval has been studied widely. News video is one of the most important media for users to get the information. It is an urgent problem to retrieve the interested news from a huge amount of news video efficiently and correctly. This paper takes news video as a research object, and discusses several problems in the process of content-based video retrieval, including shot segmentation based on cut detection, key frame selection, topic caption retrieval, news story segmentation, news video retrieval based on keywords.(1) Video segmentation is the first step, this paper offers a video shot detection method which is based on self-adapting dual-threshold compare, to complete the detection of the cut and the wipe, which achieves good experiment results. Because topic caption contains important semantic information, this paper offers a key frame selection method based on the frame of topic caption, which basically can show the content of the shot.(2) The retrieval of topic caption contains three parts. First based on analyzing of recent researches, as to the caption detection, the caption detection method is designed, which is based on texture features of 3/10 spatio-temporal slice. Secondly, the caption location method is adopted, which is based on wavelet transformation and SVM and achieves good experiment results. Thirdly, it employs the interpolation method to enhance the caption, then binary and partition the caption, and it uses the given Han Wang OCR software to recognize the characters.(3) Based on the research of news story segmentation, the method based on caption is discussed to the segmentation of news. Based on the retrieval of the caption, and the detection of slice and anchorperson shot, news video is segmented into a series of news story, and the experimental result indicate that this method is efficient. Based on the result of news story segmentation and the label of topic caption, the news retrieval based on key word is realized.更多还原