An adaptive thresholding approach for image denoising using redundant representations

A frequently used approach for denoising is the shrinkage of coefficients of the noisy signal representation in a transform domain. Although the use of shrinkage is optimal for Gaussian white noise with complete and unitary transforms, it has already been shown that shrinkage has promising results even with redundant transforms. In this paper, we propose using adaptive thresholding of redundant representations of the noisy image for image denoising. In the proposed thresholding scheme, a different threshold is used for each representation coefficient of the noisy image in an overcomplete transform. In this method, each threshold is automatically set based on statistical properties of the noise in the redundant transform domain. In our algorithm, adaptive thresholding is applied to redundant representations of noisy image blocks. Simulation results show that our method achieves the state-of-the-art denoising performance.