Contourlet Based Natural Scene Statistics Using Student'S T Distribution

In this work, we consider contourlet based image decomposition as a convenient framework for localized representation of images simultaneously in space, frequency (scale) and orientation. Since the subband marginal distributions of natural images in the contourlet domain are highly non-Gaussian with leptokurtotic behavior, we propose the Student's t probability density function (pdf) as a prior for modeling the contourlet coefficients of natural images. As an extension, we also consider the bivariate form of Student's t pdf in order to capture the across scales dependencies of the contourlet transform. We validate our proposal by adopting subjective and objective measures.