Research On Image Denoising Algorithms Based On Nonlocal Means

Image denoising plays a significant part in image processing. The noise reductioncan not only improve visual qualities but also enhance images’ post-processing. Thisthesis does lots of research on several denoising methods with emphasis on nonlocalmeans. We investigate its advantages and make some improvements on itsdisadvantages. Experimental results verify the effectiveness of the proposed methods.A new method for noise estimation is proposed. After whitening noise bynormalizing the wavelet coefficients at different bands, we can estimate the variance ofunknown noise using robust median estimator. Experimental results demonstrate itsprecision and robustness. The original nonlocal means algorithm adopts fixed globalfiltering parameters for all pixels without considering the different properties of pixels.By classifying pixels according to its edge information, we can adjust the intensity ofsmoothing adaptively for every pixel. Denoising results show that more noise isremoved and less structure is lost. Finally, a fast nonlocal means method is put forwardbased on Fast Fourier Transformation (FFT) and Summed Square Image (SSI). Thisaccelerating method is about10times faster than the original version.