Improved Bivariate Shrinkage Functions For Image Denoising

Image denoising is an important problem in image processing. The purpose of image denoising is to eliminate noise interference and improve image quality by some treatments with prior knowledge of the noise image. In this paper, we present one novel denoising algorithm based on Block-wise and another iterative denoising algorithm. Inessence, they are the extensions of bivariate shrinkage denoising.Most simple nonlinear thresholding rules for wavelet-based denoising assume that the wavelet coefficients are independent. However, wavelet coefficients of images have significant dependence, such as the dependence between the coefficients and their parents and the dependence between the coefficients in the same scale. Block-wise bivariate shrinkage functions for speckle reduction exploits not only the dependence between the coefficients and their parents but also the dependence between the coefficient and its neighbours. Where, the parent coefficient is used to determine the threshold value and take the threshold processing with the coefficients in the same scale, Experiment results demonstrate that the proposed algorithm shows great advantage in noise reduction over the original bivariate shrinkage algorithm.Iterative maximum likelihood denoising based on bivariate shrinkage, the key idea is to model wavelet transform coefficients with prior probability distributions and utilize previous estimation to refine prior probability iteratively. In theory, the corresponding bivariate MLE estimator is derived based on wavelet coefficients and previous estimation. Experiment results show that this iterative process noticeably improves the denoising performance.