The Metadata Generation System comprises three major sub-components namely the Human Annotation Tool (HAT), the Automatic Annotation and the Semi-automatic Annotation. The approach for immediate and automatic extraction and annotation of metadata from video feeds is by giving semantics to combinations of heterogeneous low-level visual features. This approach involves firstly, semantic scene classification, including key-frames extraction, similarities determination between shots, and rule based estimation of scene boundaries. Secondly, fuzzy logic based categorizing, including paradigm, Fuzzy membership function, and fuzzy feature generation and similarity measure. Thirdly, automatic sports video annotation based on robust dominant colour region detection, combined with motion feature analysis, and finally audio indexing and search tool AudiMining. The AudioMining, newly introduced to the LIVE System for searching in the audio tracks of the production archive, was specially adapted to Austrian speech. The audio tracks were analysed and the results stored for retrieval via a dedicated Web page.

OnlineAutoAnnotation Tool: High level processing is integrated with low level processing directly in video compressed domain. From the root node, which characterises the semantic type of the input video, a number of hierarchical classifications are constructed by internal nodes, and knowledge is annotated including human face detection, close-up views, outdoor scene / indoor scene, buildings, different types of sport. Below is screenshot of the Annotation Tool.