Extraction of TV news articles based on scene cut detection using DCT clustering

Y. Ariki,Y. Saito
DOI: https://doi.org/10.1109/ICIP.1996.560881
1996-09-16
Abstract:This paper describes a system which can automatically separate and extract articles included in TV news program. At first, the system detects scene cut points based on clustering of video frames compressed by discrete cosine transformation (DCT). This clustering technique is robust for abrupt changes caused by camera flushing. Then, on the detected scene cut points, the main studio frame is estimated according to a loop syntax that each article consists of several scenes starting from the main studio, followed by field reporting and ending at the main studio again. The TV news article is extracted as a frame sequence between the estimated main studio frames. As a result, we obtained 99.8% of news article extraction rate for 30 days news video database.
What problem does this paper attempt to address?