Feature Relevance Learning in Content-Based Image Retrieval Using GRA

In the uncertain and incomplete system study, the Grey Relational Analysis(GRA) method in grey system theory throws emphasis on the problem of "small-sized data samples, poor information and uncertainty" which cannot be handled by traditional statistics. As user’s query requirement may be ambiguous and subjective sometimes in content-based image retrieval, the query results are uncertain to some extent; therefore, retrieval process can be treated as a grey system, and the query vectors and the weight values of image features as the grey numbers. So, it is a good approach for us to develop a relevance feedback technique for content-based image retrieval using the GRA method in grey system theory. In this paper, we propose a novel relevance feedback technique for content-based image retrieval using the GRA method in the grey system theory. The key idea of the proposed approach is the grey relational analysis of the feature distributions of images the user has judged relevant, in order to understand what features have been taken into account (and to what extent) by the user in formulating this judgment, so that we can accentuate the influence of these features in the overall evaluation of image similarity. The proposed method, which allows the user to retrieve the image database and progressively refine system’s response to the query by indicating the degree of relevance of retrieved images, dynamically updates the query vectors and the weights for similarity measure in order to accurately represent the user’s particular information needs. Experimental results show that the proposed approach captures the user’s information needs more precisely.