Interactive content-based image retrieval using Laplacian mixture model in the wavelet domain

In this paper we demonstrate the use of the Laplacian mixture model to approximate the shape of the distribution of wavelet coefficients. The peaky nature of wavelet coefficient distributions can be modeled with only two components in the Laplacian mixture, reducing the computational complexity. The parameters of this mixture model form a low dimensional feature vector representing the texture content of the images. An interactive approach involving the radial basis function (RBF) is used to adapt the retrieval system to the user needs. Adopting the RBF as a non-linear metric to model the similarity enhances the retrieval ratio substantially. Experimental evaluation using the Brodatz image database indicates that the proposed method performs better than the conventional content based image retrieval systems both in retrieval accuracy and computational complexity.