An Efficient Approach to Wavelet Image Denoising

This paper proposed an efficient approach to orthonormal wavelet image denoising, based on minimizing the mean square error (MSE) between the clean image and the denoised one. The key point of our approach is to use the accurate, statistically unbiased, MSE estimate-Stein's unbiased risk estimate (SURE). One of the major advantages of this method is that; we don't have to deal with the noiseless image model.Since the estimate here is quadratic in the unknown weights, the problem of finding thresholding function is downgraded to solve a linear system of equations, which is obviously fast and attractive especially for large images. Experimental results on several test images are compared with the standard denoising technique Bayes Shrink, and to benchmark against the best possible performance of soft-threshold estimate, the comparison also include Oracleshrink. Results show that the proposed technique yields significantly superior image quality.