Deep-Learning based Bitstream Analytics for Value Discovery in Video

Video comprises the major communications and entertainment media asset, occupying more than 60% of todays Internet traffic. Yet, video remains the least-manageable element of the big data ecosystem. This is because all state-of-the-art methods for high-level semantic description in video require either manual annotation or compute-intensive video decoding and processing. This EPSRC-funded PhD project, together with the co-funding from The Media Institute (TMI) and its partners, aims to create a robust and performant ecosystem of software tools and infrastructure components to uniquely identify and describe video attributes within networks and file systems. This establishes a foundation for content owners and service providers to protect their video assets from piracy, measure viewer traffic, and enrich asset and rights management and recommendation services, all with substantially advanced simplicity and automation in comparison to existing methods. The project builds on novel video signature extraction technology developed by previous work of TMI and UCL. Deep learning methods will now be added and the signature extraction and content classification will be carried out using novel compressed domain information extraction. This enables semantic identification of media regardless of platform (film, television, web, OTT, mobile).