Statistical modelling of wavelet coefficients of CT scan image

Prior knowledge of wavelet coefficient statistics is a key issue in the development of better quantization strategy for enhancing compression efficiency of digital images. Since statistics of medical images are quite different from those of natural images, there is a need for statistical modelling of wavelet coefficients in different subbands. This paper examines the suitability of Student-t, Pareto Weibull and Gaussian distributions for modelling the wavelet coefficients of various subbands in a CT scan image to improve the compression efficiency. It has been found that the statistics of wavelet coefficients in the CT scan images can be better approximated by the generalized Student-t distribution for negative wavelet coefficients whereas generalized Pareto distribution provides better fit for the non-negative coefficients. The results can be potentially useful in designing adaptive quantizer for achieving improved compression gain and reducing computational complexity for medical image coders.