Concept-based image retrieval system

Abstract

Currently, most image retrieval systems rely on text annotation of pictures as the basis to index and retrieve image data. The main limitations of using free-text to describe images are the problems of completeness and consistency. Recent works have addressed the completeness problem by using the document space modification technique to update the image descriptions using the users' relevant feedback. However, the updating process can be inconsistent and that the terms used to update the image descriptions can be ambiguous. In this paper, we describe the use of a concept model of the image collection as the basis to guide the retrieval and updating of image contents. Relevance feedbacks from the users during retrieval activities are used to update the concept representations and image descriptions. The concept structure helps to ensure that only meaningful terms in the domain are used to update the image contents, and to overcome the ambiguity of the meanings of terms in a free-text environment. The system has been tested on an image database in the domain of Singapore history and has been found to be feasible.