A tree-structured image retrieval in the wavelet transform domain

In this work, we address the problem of image retrieval in the wavelet transform domain. More precisely, the wavelet coefficients are considered as realizations of a Generalized Gaussian Distribution (GGD) whose parameters are the signatures of the image. The retrieval procedure is based on a tree-structured search in the feature space that fully exploits the multiscale property of the image representation. Our main contribution consists in improving the design of such tree by employing an unsupervised classification algorithm instead of the conventional supervised K-means algorithm. Experimental results indicate significant gains in terms of precision and recall w.r.t. the K-means based design of the tree.