【作者】 陆兵；

【导师】 李士进；

【作者基本信息】 河海大学 ， 计算机应用技术， 2007， 硕士

【摘要】 文本是视频中重要的内容信息。视频中文本的检测和识别在视频分析过程中起到很大的作用。文本可以作为视频片断的内容标识和索引，例如在新闻视频中出现的新闻摘要，可以作为该段新闻内容的描述，用于新闻视频资料的检索。所以对视频文字的检测和分析是视频分析的重要内容。而检测视频中文字的出现及其准确位置，并将文字从复杂多变的背景中分割出来，是视频文字分析处理的基础。文本信息提取系统主要包括文本检测，文本定位，文本跟踪，文本提取，文本增强和OCR识别六个部分。本文重点研究了文本定位的方法，提出了一种基于投影分析与支持向量机学习相结合的文本定位方法，试验表明该方法比单纯的基于边缘的方法或是学习的方法都要好。首先采用投影分析的方法将可能的文本区域提取出来，然后再采用基于支持向量机学习的方法将提取出来的文本区域中的虚假文本区域排除掉。该方法虽然比基于边缘的方法多了一步，但文本区域的检准率有了较大的提高。与一般的基于学习的方法相比，该方法不必对整个图像区域进行特征计算，减小了计算的时间复杂度。在使用支持向量机进行文本分类时本文采用了小波，角点，扫描线和区域内边缘点的重心位置等特征。论文最后用该方法用于广告视频文本的检测，采用基于多分辨率分析的方法定位广告文本。通过比较发现，新闻中的文本出现位置比较固定而且各个电视台的文本都有各自固定的格式，但广告中的文本无论是大小，字体都是不一样的，利用这一差别可以对广告片断的起始位置有一个更加精确的定位。实验结果表明该方法可以很好的定位出广告文本。更多还原

【Abstract】 Text is part of the important information in videos. Text detection and recognition in videos can help a lot in video content analysis and understanding, since text can provide concise and direct description of the stories presented in the videos. In digital news videos, the superimposed captions usually present the involved person’s name and the summary of the news event. Hence, the recognized text can become a part of index in a video retrieval system. The importance of video can be estimated by the recognized text. So text detection and analysis is important in video analysis. Detecting the accurate position of text in the video and segmenting text from the complex background are the foundation of video text analysis.The text information extraction system can be divided into the following six parts: text detection, text localization, text tracking, text extraction, text enhancement and text recognition. This thesis focuses on the research in text localization. The projection analysis of edge based method and the learning of support vector machine based method are combined to localize text on videos. It has shown good results in the experiments compared to the simple edge based method and the learning based method. The text localization can be divided into two steps. In the first step, the potentially text area are extracted by the edge method. In the second step, support vector machine is used to classify the actual text areas and the false text areas. The false text areas are removed in this step. This method improves the precision rate of text areas. Compared to the learning based method, this method doesn’t need to compute the texture of the whole image. Instead, it only computes the texture of the text areas. This algorithm can reduce the time complexity. The textures used in the support vector machine are wavelet, corner, line and the center of gravity of the text areas.This method is applied in localizing text in advertisements. A multi-resolution based method is used to localize text in advertisements. It is a part of the advertisements detection system. It is obvious that texts in the news are more formal and its positions of texts are in a certain areas. But texts in the advertisements are different from each other in size and style. The method can give out a more accurate position of advertisements. And it has shown good results in the experiments.更多还原