VIDEO BASED OCR: A CASE STUDY OF REAL TIME IN- SCREEN SUBTITLE RECOGNITION

NPO, the Netherlands public broadcasting organization, has implemented an Optical Character Recognition (OCR) solution for real time conversion of in-vision subtitles to spoken subtitling audio. In the Netherlands, television programmes in foreign languages are subtitled. NPO offers access services for the visually impaired in the form of spoken subtitling: speech-synthesized audible text. For this to work correctly in the past, pre-produced subtitle files or embedded data within the live studio feeds were necessary. Programmes are now being contributed more and more with subtitles already present in the active video - the so-called 'title-copies' or 'burned-in subtitles' in former technical jargon. The percentage of spoken subtitled programmes has thus clearly shown a decreasing trend, just when the political call for access services for the visually impaired has been growing louder. NPO needed to find a solution and teamed up with Dutch research organisation TNO, the postal-market technology provider Prime Vision, and NPO's playout facility provider Ericsson Broadcast Services, to work towards solving this problem. With success…