Abstract Content piracy is increasing and spreading rapidly across different regions considering the kind of online infrastructure that we have these days. The primary aim of these online infrastructure is to provide the platform where-in authorized and legal content is been pushed from the service provider till the end-user. Over a period of time, the digital online infrastructure system has been misused by pirates to copy the original content and retransmit this pirated content using the same digital online infrastructure system. One of the methods to check if it is a pirated content is through visual analytics of broadcast logo. In this paper we will be discussing on the openly available datasets of TV broadcast channel logos (Indian channels), development of new scalable TV broadcast channel logo corpus across different regions and genres. A total of 450 TV broadcast channel logos have been collected of different regional language for genres like (Sports, Movies, Kids and Cartoon, Entertainment etc.) Each Logo is subjected for various data augmentation techniques to expand the logo corpus and hence strengthening the deep learning logo classification. Further this paper discusses about using this logo corpus on YOLO v2 state of art object detection algorithm and recognizes different logo classes. Experimental results are recorded for different inference logos of varying pixel context.
Read full abstract