Modelling of the video DCT coefficients

In this paper, the normal inverse Gaussian probability density function (PDF) is presented as a highly suitable prior for modelling the DCT coefficients of videos. The parameters of the proposed prior are estimated by minimizing the Kullback-Leibler divergence between the prior and the empirical PDF obtained from the DCT coefficients of digital videos. The effectiveness of the parameter estimation technique is demonstrated through Monte Carlo tests. Experiments are carried out to study the effectiveness of the proposed prior in modelling both the full-frame and block-DCT coefficients of video data, and compare it with that of the generalized Gaussian, Laplacian and Bessel K form PDFs. It is shown that in general the normal inverse Gaussian PDF is a better model than the other PDFs.