Tokyo — Hitachi’s Central Research Laboratory has reportedly developed a technology that recognizes characters in video subtitles to help rapidly search for scenes in television shows.
According to a report from the Nikkei news service, the new technology improves upon earlier subtitle-search approaches that were hampered by changes in colors or brightness in the same line or by the addition of frames or graphics.
The company said it has not decided when it will put the technology to practical use, but it plans to formally discuss the details at the 2008 IEICE (Institute of Electronics, Information and Communication Engineers) General Conference on March 18-21.
Hitachi said it tested the technology over eight hours of videos, including sports videos to search for people’s names in subtitles. As a result, a name or names were found in 329 subtitles. Specific names were correctly recognized in 93 percent of those cases.
In the experiment, Hitachi used videos encoded in MPEG-2 and used a PC to extract and recognize characters. A series of processes where characters in the subtitles of a video are recognized and compared with pre-registered names can be conducted at about 400MHz clock cycle, the company said. A main storage capacity of about 20MB is required.
Hitachi said characters in the subtitles are recognized in four steps:
· Rectangular areas including subtitles are extracted from a video.
· Character-string areas are extracted from each line in those areas.
· The character-string areas are divided into character areas, and characters are extracted from each character area.
The character strings are compared with keywords that are prepared beforehand.
