DocumentCode :
2107536
Title :
TV Commercial Detection Based on Shot Change and Text Extraction
Author :
Meng, Li ; Yong Cai ; Wang, Min ; Li, Yuanxing
Author_Institution :
Automobile Transp. Command Dept., Acad. of Mil. Transp., Tianjin, China
fYear :
2009
fDate :
17-19 Oct. 2009
Firstpage :
1
Lastpage :
5
Abstract :
A new algorithm is proposed for TV commercial detection from color video in this paper. High shot change frequency and "still" shots of the trademark information, which are two basic characteristics of TV commercial, are exploited to distinguish commercials from general programs. Our commercial retrieval system based on shot change and text detection is realized through a slide window. First, histogram difference is computed on consecutive images and then the four common shot transitions including cut, dissolve, fade in/out and wipe are detected. Our text detection algorithm based on maximum gradient difference, allows fast filtering of scan lines without text. A commercial is determined either if its shot change frequency is satisfied or if its trademark information is detected. Performance evaluation shows that our algorithm performs well with relatively high detection accuracy for different kinds of programs when we involved the text detection, which is an important contribution of this paper.
Keywords :
image colour analysis; information retrieval; television; text analysis; video signal processing; TV commercial detection; color video; commercial retrieval system; maximum gradient difference; shot change frequency; text detection; text extraction; Change detection algorithms; Data mining; Detection algorithms; Filtering; Frequency; Gunshot detection systems; Histograms; Performance evaluation; TV; Trademarks;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Image and Signal Processing, 2009. CISP '09. 2nd International Congress on
Conference_Location :
Tianjin
Print_ISBN :
978-1-4244-4129-7
Electronic_ISBN :
978-1-4244-4131-0
Type :
conf
DOI :
10.1109/CISP.2009.5302320
Filename :
5302320
Link To Document :
بازگشت