A Simple Pixel-Adaptive Bayesian Approach to Image Denoising Using Wavelet Interscale Dependency
Pei Chen, David Suter · 2002
In this paper, an approach to image denoising is proposed by utilizing the interscale dependency of wavelet coefficients. We first analyze a new property of parent/childrentype statistics in the wavelet domain. Then, a Gaussian mixture model (GMM) is employed to fit this statistical property, where the children’s variance field is estimated by a linear relation involving their parent. Lastly, MMSE estimates for the noisy wavelet coefficients are obtained. We demonstrate by experimental comparisons that this approach compares favorably with other competing approaches. In particular, though our approach does not outperform other approaches at all noise levels, there are occasions when it does outperform other approaches and where it may fall short of outperforming other approaches, the difference in performance is not large.