Bayesian color image denoising via a joint model and space projection

Abstract

As a stochastic method, the Bayesian estimation demonstrates some advantages on image denoising, such as with image noises treated as random signals. In this paper, we propose a two-stage Bayesian framework for color image denoising, utilizing the joint prior and Gamma distributions, to model the unknowns. All unknowns are estimated and updated simultaneously using evidence analysis within the Bayesian framework. We also propose an optimal luminance/color-difference space projection for the two-stage Bayesian framework, exploiting strong correlation in high-frequency contents of different color components to improve denoising performance. Experimental results confirm that the proposed algorithm offers superior denoising performance compared with existing solutions, both from peak signal-to-noise ratio and visual quality perspectives. By comparing experimentally the performances of the proposed algorithm in different color spaces, we have proven the effectiveness of space projection in improving the image denoising.