Dynamic Object-oriented Bayesian Networks for Flexible Resource-aware Content-based Indexing of Media Streams

To effectively utilize the growing number of digital media sources in todays electronic world, easy to use computational flexible content-based access is a necessity. In contrast, content-based access must often be based on interpreting results from low-level computationally expensive feature extraction algorithms. We propose a new approach based on dynamic object-oriented Bayesian networks for encapsulating low-level feature extraction algorithms in integration with probabilistic interpretation models, within combinable high-level domain concepts. This approach opens up for content-based access in terms of high-level domain concepts at the end-user level. Furthermore, we suggest how the feature extraction algorithms can be activated on a per need basis (hypothesis driven activation). This opens up for trading off feature extraction resource usage against the reliability of the content-based access, which in turn may allow real-time content-based access in a greater range of processing environments.