Abstract: This paper presents a novel effective method of topic caption text for news video.The change of the local frame difference between adjacent frames is computed by quantified spatial brightness of pixels in order to identify the frames that caption appears and disappears.And four rules are established to distinguish topic caption from non-topic caption.Afterwards，multi-frame integration method is used to obtain one optimal topic caption detection image on the basis of the color of caption text.Finally，the Chinese character is segmented by improved projection approach.The experimental result shows that the proposed method can effectively detect topic caption for structuring and retrieving news video.

Key words: news video, topic caption, the change of the local frame, text detection, Chinese character segmentation

摘要： 提出一种新的有效的新闻视频主题字幕检测方法。利用像素的空间量化亮度值计算局部帧差变化，以便检测新闻视频字幕的出现帧和消失帧，并建立4条规则来进一步区分主题字幕和非主题字幕，然后对同一主题字幕所在帧利用多帧结合的方法生成两个候选的主题字幕文字检测帧，选择其中之一进行文字区域的定位和提取；最后用投影法分割字符，并对投影法产生的错误分割字符的常见情况，提出相应的解决方法。实验结果表明了该方法的有效性。

关键词: 新闻视频, 主题字幕, 局部帧差变化, 文字检测, 字符分割