Wavelet Denoising Based On Non-local Filter

Image denoising is a important part in the field of the image processing.The traditional image denoising method with a loss of details information for price to filter the noise of the image. How can filter the noise of the image and efficiently keep image structure details have been the research key of many scholars.Nonlocal means method is put forward in the last few years.It is different from traditional filter method. It fully utilize the redundant information of image to filter out the noise. Nonlocal average algorithm can reserves the detail of image better and effectively avoid the artifacts appeared in traditional adjacent domain filtering method.Because of these advantages, more people concern in this method, domestic and foreign scholars have proposed a lot of improved methods. Wavelet denoising based on nonlocal average is one of them. Because in wavelet domain of the original image the processing time is only in the quarter, this method can get better effect of filter and effectively reduce the processing time. But in emulation we found that this method in visual effects and filtering effect or not very satisfactory.For above those shortcomings, this paper puts forward a improved denoising method based on gereralized non-local means filter. Firstly, in order to segment the noised image and keep the ridge, we applyed the watershed segmentation based on gradient method. Then the noised image had been multidimensionally wavelet decomposed. Estimated the scale coefficients and shape coefficients of the decomposed coefficient of each layer. And for filtering each layer of the decomposed coefficients,a similar judgement has been taken and we used non-local means filtering method based on Generalized Gaussian Model before wavelet reconstruction;Finally, the processed noised image pixels which corresponding the watershed ridge replaced the corresponding pixels in the wavelet reconstructed image. Simulation results showed that this method can obtain better denoising effect than the method of non-local means filtering method based on Generalized Gaussian Model.