Instantaneously responsive subtitle localization and classification for TV applications

This paper presents an algorithm for localization and classification of subtitles in TV videos. We extend an existing static-region detector with object-based adaptive filtering and binary classification of subtitle bounding boxes, using geometry and text-stroke alignment features. Compared to this static-region detector, we reduce the number of falsely detected subtitle pixels by a factor of 20, at the cost of only 2% of subtitle pixels.