Analysis Of Video Text In Content-Based Video Retrieval

The content of texts in digital videos provides valuable information for people to understand the semantic content of videos. Specially for news videos, texts in videos include the main idea of news. Once texts in videos are detected, located and recognized, the results can be quite useful for indexing the content of videos automatically and helping people find the video chip they need by the way of text query. How to detect the texts and recognize them in video is the key problem of analysis of video text. This paper describes our work on video decoding and analysis of video text in content-based video retrieval.One aspect of our work is to produce two schemes of video decoders based the mpeg4ip according to video decoder's requirement of systems of content-based video retrieval, after analyzing new feature of H.264 and its decoding technology. Then we discuss how to realize two scheme and compare the decoding performance of two scheme.The Other aspect of our work is to analyze the image characteristics of video text. For texts in news video are almost stable, we produce the method of news local-caption extraction based on spatio-temporal distribution feature. In the algorithm's realization, two kinds of operators are used and the computing results of two operators are compared and analyzed. The algorithm uses Multiple Frame Integration(MFI) to enhance image's effect and improve the quality of texts detecting. At Last, we show the algorithm's experimental results of text recognition using business's OCR software.