Multivariate indexing of multichannel images based on the copula theory

In this work, we address the problem of multichannel image retrieval in the compressed wavelet-based domain. A wavelet transform is applied to each component. Then, two approaches are applied to extract features from the multiresolution representations. In the first one, the wavelet coefficients of each component are considered as mutually independent and hence, features are separately computed. In the second, the distribution of the multivariate wavelet coefficients is modeled by a multivariate model driven by the copulas theory. The parameters of this multivariate distribution are chosen as relevant signatures of the image. The contribution of this paper is to investigate the influence of the copula density on the retrieval performances. To this respect, we have tested the Gaussian copula density and two Archimedean copula densities (the Clayton and the Gumbel copulas). Experimental results indicate that the Archimedea n copulas outperfom the Gaussian one in terms of precision and recall of retrieval.