Image retrieval using multiple features often uses explicit weights that represent the importance of the features in their similarity metrics. In this paper, a novel retrieval method based on Bayesian Learning is presented. Instead of giving every feature a weight explicitly, the importance of a feature is regulated implicitly by learning a user's perception. Thus, the process of feature combination is adaptive and approximate to a user's perception. Experimental results demonstrate the signicance of this method for improving the retrieval efficiency.
Available at: http://works.bepress.com/lwang/18/