Semantic retrieval of multimedia by concept languages: treating semantic concepts like words

This paper presents a retrieval approach that uses concept languages to deal with nonverbally expressed information in multimedia. In operation, a finite number of elecepts of a concept language are identified and used to index multimedia documents. The elecepts then allow a large number of compound semantic queries to be expressed and operated as sentences of elecepts. We believe that managing semantics by concept languages is a prudent proposition. Not only does it extend an intuitive query regime where semantic queries can be specified more expressively and extensively, the approach also allows concept detection to be restricted to a more manageable sum of semantic classes. Two example applications are discussed in the paper. One uses a color artistry language to retrieve art images and the other utilizes a tennis concept language to retrieve tennis videos