The DIVAS proposal tackles the design and implementation of a multimedia search engine based on advanced direct video and audio search algorithms and efficient content representation in compact search databases. The driving force is to disjoint audio/video search from the availability of laboriously annotated metadata databases, while at the same time reducing significantly storage requirements and search time.
The developed algorithms will provide an alternative and complementary path to semantic metadata based audio/video content search. The proposed approach advocates the automatic extraction of content features (fingerprints), directly from the compressed content, thus greatly accelerating the fingerprint extraction phase as well as the search process.
The search databases will comprise features that uniquely characterise compressed content, and will be rather compact and suitable for binary search techniques. The search engine will be applicable to a number of different use cases ranging from "similarity" searches for audio/video web content, to information harvesting and data mining. Another interesting application is the development on "stream searchers" suitable for DRM resolution, advertisement time tracking etc.
The proposed improvements will ultimately lead to the introduction and seamless integration of audiovisual searching to ANY WEB search engine, and the location of video content ANYWHERE, irrespective of transformations and annotations, thus adding true multimedia search capability to ambient intelligence. The project will deliver complete system architecture, a system level demonstrator (available through the web for user evaluation), studies and design methodologies for application integration.
