Texture similarity measurement using Kullback-Leibler distance on wavelet subbands

The focus of this work is on using texture information for searching, browsing and retrieving images from a large database. In the wavelet approaches, texture is characterized by its energy distribution in the decomposed subbands. However it is unclear on how to define similarity functions on extracted features; usually simple norm-based distances together with heuristic normalization are employed. We develop a novel wavelet-based texture retrieval method that is based on the modeling of the marginal distribution of wavelet coefficients using the generalized Gaussian density (GGD) and a closed form Kullback-Leibler distance between the GGDs. The proposed method provides greater accuracy and flexibility in capturing texture information while its simplified form has close resemblance with existing methods. Experimental results indicate that the new method significantly improves retrieval rates, e.g. from 65% to 77%, against traditional approaches while it has comparable levels of computational complexity