A Research On Image Denoising By Non-Local Means

This thesis mainly discusses about image denoising, which has disturbed researchers for quite a long period. It does researches on Non-Local algorithm and its application in image denoising. The thesis emphasizes on the following parts: implementation on Non-Local means, comparison of method noise among Non-Local means and other filters, rational selection of neighborhoods, and Non-Local means algorithm based on General Gaussian Distribution.The noise reduction will affect the whole work of image processing. It's extremely difficult to distinguish unknown noise from details and structures in natural images. The basic idea of denoising is average, so the key point is how to do smoothing while preserving details or high frequency parts. A. Bades, et al brought forward the concept of method noise, which changed the viewpoint of the problem.Based on the above, the contributions of our work mainly focus on the following aspects:1. Formulae of method noise for the Gaussian smoothing filter, the anisotropic filter, the Wiener filter, the translation invariant wavelet thresholding, and Non-Local means algorithm are deduced. The experiments'result shows that Non-Local means is better than any mentioned filters.2. In order to accelerate the Non-Local means algorithm, filters that eliminate unrelated neighborhoods from the weighted average are introduced. These filters are based on local average gray values and gradients, pre-classifying neighborhoods and thereby reducing the original...