Semantic Retrieval of Multimedia by Concept Languages
Jose A. Lay, Ling Guan · 2006
[Treating semantic concepts like words] Despite the quantum leap in computing performance, semantic retrieval of multimedia remains an unsolved problem. At the heart of the challenge is managing semantics. A retrieval engine works to bring together a set of documents that are relevant to a query. However, humans are endowed with such sophisticated relational competence that we effortlessly relate a concept to a great many others. For example, when we search for rose, we may relate it with a prickly fragrant flower, a person named Rose, or equally, a rock band called Guns N ’ Roses. Consequently, retrieving multimedia requires a means by which semantics can be specified and a mechanism through which relevant documents can be brought together. In this article, we describe a retrieval approach based on concept languages. We show how the latter can be used as a compelling means to express semantics and demonstrate how it can serve as an efficient mechanism to collocate documents. 1053-5888/06/$20.00©2006IEEE IEEE SIGNAL PROCESSING MAGAZINE [115] MARCH 2006EARLIER WORKS In content-based retrieval (CBR), the main approach has so far centered on the use of perceptual features—i.e., audio and visual descriptors extracted from a document [1], [2]. At first,