Using Context Information to Improve Retrieval Accuracy in Content-Based Image Retrieval Systems

Current image retrieval techniques have shortcomings that make it difficult to search for images based on a semantic understanding of what the image is about. Since an image is normally associated with multiple contexts (e.g. when and where a picture was taken,) the knowledge of these contexts can enhance the quantity of semantic understanding of an image. In this paper, we present a context-aware image retrieval system, which uses the context information to infer a kind of metadata for the captured images as well as images in different collections and databases. Experimental results show that using these kinds of information can not only significantly increase the retrieval accuracy in conventional content-based image retrieval systems but decrease the problems arise by manual annotation in text-based image retrieval systems as well. Keyword : Image Retrieval, Context-Aware Computation, Annotation