Multimedia Ontologies


In contrast to ontologies of other knowledge domains, video ontologies need a specific set of motion events to represent spatiotemporal changes of video scenes, which are characterized by subconcepts, multiple interpretations, and ambiguity. Research results in structured video annotations are particularly promising for constrained videos, where the knowledge domain is known, such as medical videos, news videos, tennis videos, and soccer videos. In spite of the benefits of multimedia reasoning in video scene interpretation and understanding, most multimedia ontologies lack the expressivity and mathematical constructs necessary for complex reasoning tasks.

After recognizing the limitations of previously released multimedia ontologies, many of which are actually knowledge bases only rather than fully featured ontologies, Dr. Sikos has developed a novel ontology, VidOnt, which is the most expressive multimedia ontology to date, formally grounded in the decidable SROIQ(D) description logic and complemented by a role box and a DL-safe ruleset unmatched in multimedia ontology engineering. Specifically designed for video representations, VidOnt covers the professional video production and distribution domains.

Dr. Sikos is also working on the semantic representation of 3D scenes and objects, such as cultural heritage artifact models and 3D anatomical models created in AutoDesk 3ds Max, AC3D, Modo, Blender, or Seamless3d.