High-level feature extraction is to automatically obtain possibly human-recognizable features beyond the level of machine-recognizable ones such as keywords in text and speech, and color and texture in image and video. Some desirable human-recognizable features from video are moving and still objects from video; concepts from spoken language and characters on images; and people, places, and organization mentioned in the content of video. They provide rich foundation for synthesizing information in semantic level.

Content-based presentation relies on clustering and context analysis techniques over an entire set or some selected subsets of information, which is represented by high-level features. While clustering focuses on grouping similar logical segments, context analysis emphasizes discovery of relationships between concepts in logical segments. Both techniques provide the capability of conceptual integration in addition to temporal and geospatial integration of information.