Multivariate statistical modeling of images with the curvelet transform

This paper proposes a theoretical and statistical study to characterize the dependencies of the curvelet coefficients of images across position, scale and orientation. Our study was based on estimated histograms of the marginal and joint distributions to study the statistical properties of curvelet coefficients, and on the mutual information to measure the level of dependence between these coefficients. Finally, a novel multivariate statistical model, namely the anisotropic multivariate generalized gaussian (AMGGD), was proposed to characterize these dependencies.