A Video Caption Extraction Approach Based On Character Segment And Unsupervised Clustering

Caption is an important part of the video information. It not only contains an abundant high-level semantic, but also is always used to mark the video. Moreover, it is a vital clue to help people to understand the video. So retrieving and extracting the captions in the video has been a critical part of video parsing. Among them, how to split the caption from complex background in the video, and make sure the OCR could recognize it correctly, has been the key technology of video processing. This is the main problem that this paper tries to solve.Through the study and analysis of previous related works, considering the problems of complex background and strong noise, we propose a video caption (?)extraction approach based on character segment and unsupervised clustering, this method segments the characters by using its gradient feature, which solve problems caused by mutual interference and global noise. Then, according to the color and location information, we cluster the character image and separate the text layer as soon as possible. Finally, using connected component analysis and other methods we reduce the interference noise and receive a clear binary image to be identified.Experimental results show that the proposed method can effectively remove the background image of the caption area and increase the recognition rate. Moreover, this approach has been successfully applied in the video caption extraction system and makes the whole system achieving the goal that automatically extracts captions from video. It also has strong robustness.