Control structures for incorporating picture-specific context in image interpretation

摘要: This paper describes an efficient control mechanism for incorporating picture-specific context in the task of image interpretation. Although other knowledge-based vision systems use general domain reducing computational burden interpretation, to our knowledge, this is first effort exploring collateral information. We assume that constraints on picture are generated from a natural language understanding module which processes descriptive text accompanying pictures. have developed unified framework exploiting these both object location and identification (labeling) stage. In particular, we describe technique constrained search context-based vision. Finally, demonstrate effectiveness approach PICTION, system uses captions label human faces newspaper photographs.