Image Denoising using Multi-Resolution Coefficient Support Based Empirical Wiener Filtering

In this paper, a new image denoising algorithm using simple thresholding operations and wavelet coefficient magnitude based Wiener filtering is proposed. A hard thresholding operation is initially performed on the noisy wavelet coefficients. The initial significance map is then refined by use of multi-resolution coefficient support map which considers the local spatial features of the image. As a final denoising step, optimal Wiener filtering is performed on the thresholded wavelet coefficients using only magnitude information. The performance of proposed algorithm is evaluated on standard test images and found to perform competitively to the state-of-art image denoising algorithms in the literature.

1. INTRODUCTION

The wavelet transform of images provide a compact distribution of energy within a few, large set of wavelet coefficients, these large coefficients are generally spaced around the local spatial features of the image. Due to its sparse representation ability simple thresholding operations are proven to be asymptotically optimal for removing additive noises present in the image observations [1], [2]. Inspite of effective decorrelation, wavelet coefficients exhibit strong dependencies along similar spatial and scale positions. These spatial dependencies are exploited in the more recent denoising algorithms by use of some statistical models. In [3], [4] wavelet coefficients are modelled as Gaussian distributions with locally varying variances, and the variances of the coefficients are estimated from the neighborhood coefficients, and these methods essentially differ in how they estimate the local variances. The techniques in [5], [6], [7] calculate Lipschitz exponents to create an initial map indicating edges, the initial map is further refined by the use of Markov random field models [6], [7].