MMSE based map estimation for image denoising

Denoising of a natural image corrupted by the additive white Gaussian noise (AWGN) is a classical problem in image processing. The NeighShrink [17,18], LAWML [19], BiShrink [20,211, IIDMWT [23], IAWDMNC [25], and GIDMNWC [24] denoising algorithms remove the noise from the noisy wavelet coefficients using thresholding by retaining only the large coefficients and setting the remaining to zero. Generally the threshold depends mainly on the variance, image size, and image decomposition levels. The performances of these methods are not very effective as they are not spatially adaptive i.e., the parameters considered are not smoothly varied in the neighborhood window. Our proposed method overcomes this weakness by using minimum mean square error (MMSE) based maximum a posterior (MAP) estimation. In this paper, we modify the parameters such as variance of the classical MMSE estimator in the neighborhood window of the noisy wavelet coefficients to remove the noise effectively. We demonstrate experimentally that our method outperforms the NeighShrink, LAWML, BiShrink, IIDMWT, IAWDMNC, and GIDMNWC methods in terms of the peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM). It is more effective particularly for the highly corrupted natural images. (C) 2013 Elsevier Ltd. All rights reserved.