Digital Video Subtitles In The Detection And Extraction

Detecting and extracting the caption presents in the digital video plays an important role in the caption-based video retrieval. Using caption to retrieve the video can reduce the complexity of video retrieval and increase the speed and veracity of video retrieval. So detecting and extracting the caption from video automatically is meaningful for content-based video retrieval. The caption is embedded in every frame. In order to extract the caption, at first, segmenting the video into shots, and extracting the frame which contained the caption from every shot, then locating the caption from the caption-contained frame. At last, extracting the caption and using the given OCR software to recognize the characters.This paper's primary job as follows:1. In video segmentation part, this paper improves the dual-threshold compare method, makes the threshold to select automatically. Through the software programming, this method has detected the cut and wipe shot. Lots of experiments show that this method is practical.2. In caption-frame detection part, this paper improves the content-based analysis of key-frame extraction algorithm, combines the key-frame extraction and the caption-contained frame detection, selects one key-frame which has the maximum brightness variance as caption frame. This improved algorithm is mainly aimed at the news video, has specific aim, but lacks of generality.3. In caption location part, this paper puts forward an algorithm based on wavelet transform. At first, graying the caption frame and reconstructing the gray-scale frame based on wavelet weighted reconstruction, using binary morphological processing to remove non-caption block, then applying the multi-scale edge detection to enhance the edge which improves the recall-precision, finally locating the caption area through projection method. Simulation experiments show that it is insensitive to threshold, and even in complex background, still has better effect.4. In caption recognition part, first, improving resolution on interpolation method, using global threshold method to process the caption, then segmenting caption based on projection method, finally using OCR software to recognize the characters. Experiments show that this method is satisfying.