Pic-A-Topic: efficient viewing of informative TV contents on travel, cooking, food and more
Large Scale Semantic Access to Content (Text, Image, Video, and Sound)
| Hi-index | 0.00 |
Recently, due to the rapid spread of digital video recorders capable of storing a large number of TV programs, there is a great need for easy interfaces for viewing program content and for accessing scenes of interest. To provide such functions, it is necessary and effective to segment a recorded TV program into temporal sections according to the content structure. However, in conventional approaches such as cut-detection methods, content is chopped up into segments that are too short, and the computational costs of intelligent recognition methods for speech, face, and text are prohibitive in many cases. The author has developed a light and high-speed algorithm which estimates the structure of video content and completes temporal segmentation instantly after the recording. The proposed method introduces a measurement which denotes the likelihood of a given temporal part of a recorded TV program being a dialog scene, according to the appearances of similar shots. With this measurement, related consecutive shots are determined as being in a dialog scene, and consequently topics in a news program and sections in a variety show are effectively estimated. In this paper, the author discusses development, implementation, and effectiveness of the system which provides automatic segmentation of a variety of recorded TV programs, applying the proposed method. © 2006 Wiley Periodicals, Inc. Syst Comp Jpn, 37(6): 44–54, 2006; Published online in Wiley InterScience (). DOI 10.1002/scj.20438