IST project develops system for archiving and retrieving audiovisual content
As increasing amounts of information are stored in video, picture and audio formats, the challenge of efficiently searching and retrieving audiovisual content becomes ever more pressing. One solution may have been found in the shape of a recently concluded EU research project. Supported by one million euro from the Fifth Framework Programme's IST programme, the combined image and word spotting (CIMWOS) project has developed a new system for the classification and retrieval of multimedia content. The first challenge for the CIMWOS consortium was the development of search algorithms specifically designed for classifying audiovisual material. For audio content, this requires the use of continuous speech recognition technologies in order to produce transcripts of what is said by whom, and about which topic. The system's image processing algorithms, meanwhile, extract details such as scenes, logos, objects and particularly faces, as well as the captions and subtitles that may accompany them, to act as the basis of future searches. The CIMWOS consortium includes partners from Belgium, France, Austria, and Switzerland, and was coordinated by Dr. Athanassios Protopapas from the Institute for language and speech processing in Athens, Greece. Dr Protopapas told CORDIS News: 'Summarisation of material is very important for skimming over large segments of audiovisual content without having to watch every moment. CIMWOS is a great step ahead of previous research projects because it incorporates very advanced processing technology.' Having succeeded in finding a method to summarise and classify information, the next challenge was to develop a user friendly interface to efficiently search and retrieve it. A database of broadcast news items and documentaries in three EU languages (English, Greek and French) was used as the basis for such an interface, and the team devised a web based system that allows users to perform searches based on media descriptions, content metadata, and free text. Furthermore, because high quality video can take up a large amount of memory and bandwidth, the CIMWOS team developed a system that allows users to search large audiovisual databases from their workstations, and delivers the results in a compact preview format. Having selected which results are relevant, the user need only download these, saving both time and infrastructure costs. Dr Protopapas believes that CIMWOS will be an important resource for promoting the reuse of existing audiovisual content, thus reducing budgets for new productions. Furthermore, he hopes that the methodologies developed by the consortium will promote the standardisation and unification of audiovisual databases, paving the way for large scale European digital libraries. While the consortium accepts that some further developments are still needed, they stress that the system is near a practically usable state, and that it offers features not yet available in commercial applications. Indeed, as one expert external reviewer concluded: 'In the archive domain, the BBC has a standard for indexing the in house production: eight hours of a documenter for each hour of a produced programme. With the help of CIMWOS tools, I believe that the work could be reduced to four hours. I think that this could be enough to convince a general manager with 120,000 hours [of programmes] to digitise that this project is of strategic importance.'