Image Denoising with Trivariate Shrinkage Based on Sharp Frequency Localization Contourlet

Conventional thresholding shrinkage for denoising is designed with assumption that the coefficients in transformation domain are independent. However, in practice, natural images’ coefficients in transformation domain have significant dependencies. In this paper, we proposed a novel method for image denoising by exploring the dependencies among the coefficients. The method considered three corresponding coefficients, including the noisy coefficient, its parent coefficient and its neighbor coefficient based on the Sharp Frequency Localization Contourlet, and established a trivariate distribution model to estimate the latent coefficient. Furthermore, the shrinkage function in model is derived under the Bayesian framework. The experiment results showed that the performance of proposed method outperformed the current denoising method.