Semantic representation of text captions to aid sport image retrieval

Image captions represent manual semantic annotation of images. These act as essential cues to represent the semantics of an image. This paper describes the process of representing, discovering, storing the semantics in a knowledge base, and then applying the semantics to aid the retrieval of visual information. We exploit a Natural Language Processing (NLP) framework in order to extract the knowledge from image captions and to transform those unstructured data into a semantic model. The novelty of the proposed framework is to use a semantic model to find implicit relationships among the concepts of photographs which are not mentioned directly in text captions. Latent Semantic Indexing (LSI) is deployed to handle ontology imperfections. Experiments tested and validated the major hypotheses of this approach.