基于灰度差分和FCM的新闻字幕检测及提取

Extraction and Detection of News Captions Based on Gray-Scale Difference and Fuzzy C-Means

新闻视频字幕是新闻内容的高度概括，检测及提取出视频字幕对视频的语义化具有极大意义。通常视频的前后画面具有相关性，新闻字幕出现和消失会引起较大的帧间差，计算帧差图像较边缘检测具有运算简单、速度更快的优点。由此，提出一种高效算法，该方法首先计算相应视频灰度差分图像，然后通过水平和垂直方向投影门函数进行初步判定，再结合区域几何特性排除干扰，最后用模糊C均值聚类对所得的彩色字幕块进行分割，实验结果表明，该方法可以获得较好的字幕检测和提取效果。

The retrieval of a news title caption is of very important significance for video semantic because the caption is the high-level overview of the news. Generally, there is some correlation in the scene between adjacent frames, then a big change of the gray-scale difference is made among these adjacent frames if caption text is appeared or diappeared. Calculating gray-scale difference is much easier and faster than edge detection. An algorithm is presented in this paper which first calculates gray-scale difference statistics of the adjacent frames , secondly assumably locates the caption＇s region with the gatefunction in horizontal and vertical projection respectively , and then eliminates some false region by geometrical characteristics. Finally, the obtained colorful character area is segmented with Fuzzy C- Means. The experiment results demonstrate that this algorithm is effective and more precise.