视频文本提取技研究及在虚拟卡拉ok中的应用.pdfVIP

摘 要 摘 要 数字视频在教育、娱乐及其它多媒体应用中发挥着重要作用。为了满足需求，各种 各样的视频分析、检索技术被提出，其中，视频文本包含了丰富的高层语义信息，可用 来对视频进行语义标注。 本文研究从视频文件中提取出字幕文本的方法。首先从视频文件中自动截取若干视 频帧，对每幅视频帧采用基于 Sobel 算子的最大梯度算法检测文本边缘；再利用文字丰 富的边缘信息，使用方差投影定位文本区域；然后通过计算区域边缘密度过滤虚假文本 区域；最后利用数学形态学处理文字边缘图像，通过计算颜色标准差获得文本的颜色信 息，从而分割出文本图像。实验表明，本文提出的方法不但可以比较准确地定位视频文 本，过滤虚假文本区域，还可以有效去除文字区域的背景图像，提高文字识别率。 关键词 视频图像 视频检索 文本提取 数学形态学 I Abstract Abstract Digital videos play an important role in education, entertainment and other multimedia applications. To meet the demand, various techniques of video analysis and search have been proposed. The video text contains plenty of high-level semantic information which can be used in semantic annotation of video. This paper studies the methods of locating and segmenting subtitle text from the video. First of all, intercepting a number of video frames from the video automatically, and using maximum gradient algorithm which based on sobel operator to detect text edges from each video frame. Then using the rich information of text edge and the variance projection to locate text regions. Afterwards, filtrating false text zones by calculating the edge density of the zones. Finally, mathematical morphology is employed to treat the image of text edges, and color information is obtained by calculating the standard deviation of color. Thereby, text image can be extracted from the video frame. Experiment shows that the method we proposed is not only accurately locate the video text and filtrate false text zones, but also effectively remove the background of text zones. Moreover, it can improve the rate of character recognition. Key