Robust noise estimation based on noise injection

Abstract

Noise estimation is an important premise for image denoising and the related research therefore has drawn increasing attention and interest. Recent studies show that the distribution mode of local variances in natural image can be used as a simple yet efficacious estimator of the additive noise variance, no matter what distribution the noise follows. However, this type of method has the limitation that the target image must have a sufficiently large area with low pixel value variations. Furthermore, this type of noise estimator almost always lead to overestimation without taking into account the mode of local variance distribution of the noise-free image in textural regions. To improve the accuracy of distribution-mode analysis type of noise estimation and to resolve the problem of overestimation, we propose a novel algorithm using a cascade of wavelet sub-band estimation and noise-injection based rectification. The proposed algorithm reduces the detrimental influence of textural image area, and therefore alleviating overestimation of the noise variance. Extensive experiments and comparative study show the reliability and superiority the proposed method over some existing competitors. © 2012 Springer-Verlag.