Undecimated wavelet-based Bayesian denoising in mixed Poisson-Gaussian noise with application on medical and biological images

摘要

Due to photon and readout noise biomedical images are generally contaminated by a mixed Poisson-Gaussian noise. In this paper, we propose a Bayesian image denoising methodology for images corrupted by a mixed Poisson-Gaussian noise. The proposed method first applies a Generalized Anscombe transform in order to convert the Poisson noise into Gaussian one. The PCM SαS Bayesian estimator using the undecimated wavelet transform is then performed to remove the Gaussian noise. Finally, the exact unbiased inverse of the Generalized Anscombe transformation is applied to improve the recovery of the estimated denoised image. The experiments on real medical and biological images show that the proposed approach outperforms the MS-VST method especially in the presence of a high Poisson-Gaussian noise. It also ensures a good compromise between the noise rejection and the conservation of fine details in the estimated denoised image.