Description

Automatic identification of commercial blocks in news videos finds a lot of applications in the domain of television broadcast analysis and monitoring. Commercials occupy almost 40-60% of total air time. Manual segmentation of commercials from thousands of TV news channels is time consuming, and economically infeasible hence prompts the need for machine learning based Method. Classifying TV News commercials is a semantic video classification problem. TV News commercials on particular news channel are combinations of video shots uniquely characterized by audio-visual presentation. Hence various audio visual features extracted from video shots are widely used for TV commercial classification. Indian News channels do not follow any particular news presentation format, have large variability and dynamic nature presenting a challenging machine learning problem. Features from 150 Hours of broadcast news videos from 5 different ( 3 Indian and 2 International News channels) news channels. Viz. CNNIBN, NDTV 24X7, TIMESNOW, BBC and CNN are presented in this dataset. Videos are recorded at resolution of 720 X 576 at 25 fps using a DVR and set top box. 3 Indian channels are recorded concurrently while 2 International are recorded together. Feature file preserves the order of occurrence of shots.

Related datasets