IMAGE DENOISING BASED ON GAUSSIAN AND NON-GAUSSIAN ASSUMPTION

Abstract

In this paper, we first present an adaptive intra-scale noise removal scheme, and estimate clean wavelet coefficients using new prior information with Bayesian estimation techniques. A new model using the non-informative improper Jeffreys' prior is given under the supposed Gaussian distribution for orthogonal wavelet transformation. Then, we propose a computationally feasible adaptive noise smoothing algorithm that considers the dependency characteristics of images. The wavelet coefficients are assumed to be non-Gaussian random variables for non-orthogonal redundancy transformation. The variances of the wavelet coefficients are estimated locally by a centered square-shaped window for every pixel within each subband. The experimental results show that the orthogonal wavelet transformation provides better results at the Gaussian assumption, while the non-orthogonal redundancy wavelet transformation performance tends to increase when the non-Gaussian bivariate distribution is used.