A novel relevance feedback technique in image retrieval

Abstract

The relevance feedback based approach to image retrieval has been an active research direction in the past few years. Many parameter estimation techniques have been proposed for relevance feedback. However, most of them are either based on ad-hoc heuristics or only partial solutions. In this paper, we introduce the first technique that not only has a solid theoretical framework but also takes into account the multi-level image content model. This technique formulates a vigorous optimization problem. By using Lagrange multipliers, we have derived the explicit optimal solutions for both the query vectors and the weights associated with the two-level image model. Experimental results on realworld image collections have shown the effectiveness and robustness of our proposed algorithm.