复杂背景下的视频文本定位与分割-信号与信息处理专业论文.docx

Subject ：The Complex Background of Video Text Localization and Segmentation Specialty ：Signal and Information Processing Name ：Jiaxin Li （Signature） Instructor：Yanhai Wu （Signature） ABSTRACT Text information in the video is an important clue for helping to understand video content, extracting the text information effectively is a key technology for application of automatic understanding of video content and retrieval. Commercial OCR also can not effectively identify the video text images with background is complex and strong interference, looking for a universal method of text positioning and segmentation to improve the efficiency of identification is one of the hot issues in the present study. This paper focuses on the method of video text localization and segmentation. According to the characteristics of the video text under complex background, text positioning method of corner extraction combined with morphological is adopted, aiming at the problems of different size of characters, the multi-scale corner detection method is proposed, combined with characteristics of the characters with the same color, the method of color clustering on the text area is used to get a more precise positioning; utilizing the characteristics of the wavelet-domain Hidden Markov Tree models widely used, the character and background located in the text area is modeled in the wavelet-domain, combining with the scale fusion theory and making full use of the class logo features on different scales, the problem of segmentation of the characters in the text is solved better, and images can be get. The experimental results show that the improved text positioning algorithm can effectively detect text messages of complex video, not only detect the text messages of ball game scene and video of news type, but also can be used in natural scene of the text of the detection positioning, The segmentation results of the wavelet domain HMT model is better than the Otsu. In maximum keep text informa