| Feature learning for knowledge discovery of multimedia data is a difficult yet urgent topic. The extraction of new facts from large repositories is known under various names: data mining, e-learning, case-based learning, ontology building, and knowledge elucidation. The list of names indicates there is no one preferred technique to solve the access to the content of multimedia data. The topic is relevant for archiving, publishing, content owning, I-services, e-culture, and many other application fields of future computer systems. For the purpose of discovery we have to prepare ourselves using data sets of 10,000 - 100,000 pictures, text items, audio tracks and pieces of music. Digital video produces 10 Mbyte of data per second per channel. So, a persistent bottleneck for multimedia discovery is compute power. Therefore, we aim at profiting from the opportunity of statistical methods and as a challenge search for computationally efficient solutions. Due to the methodological nature of the project, work packages will produce a software library consolidated by the software engineer supplied by Project M. - Available demonstrators: * Browsing: pictures in stock and news; * Browsing: picture information on the internet; * Compression: sound and speech encoding; * Detection: smart cameras for inspection; * Tracking: with moving camera; * Video: learning of shot location; * Tracking: the object tracker. |
