A SIZE-INDEPENDENT CAPTION EXTRACTION METHOD FOR KOREAN CAPTION WITH DECISION TREE AND THE COMPUTER READABLE MEDIUM THEREOF

摘要

PURPOSE: A method for extracting various sized Korean captions and a recording medium recording the same are provided to extract captions by determining whether connection objects are a caption or a background. CONSTITUTION: A caption candidate connection object image is produced by extracting a connection object from edges connected to a caption edge candidate image(S40). The pixel rate of the directional edges and the brightness variation of center pixels of the connection object are calculated. According to the decision tree, the connection object is adjudged as a caption or a background and a caption search result image is produced. A preprocessed image corresponding to connection object belonging to the caption search result image is binarized into the caption and the background and the binarized caption image is extracted(S50).