Image Denoising Based On Edge Preserving

Image denoising, an important branch in digital image processing, attracts many investigators to research and explore. And many image denoising algorithms, such as local neighborhood average denoising method, the denoising method based on partial differential equation, non-local means algorithm and sparse representation denoising method, are proposed. These methods show good results in image denoising, but cannot protect the image edge information well and avoid pseudo texture phenomenon. Aiming at this problem, this paper, based on existing image denoising strategies, to maintain the image edge and eliminate pseudo texture, improves partial differential method, nonlocal average method and sparse representation method used in image denoising. And the new method can remove the noise and properly preserve the edge details as well. The main work and innovations of this paper are as follows:1. The gradient threshold used to judge the flat and edge parts in partial differential denoising method is improved and a new gradient threshold function based on number of diffusion is designed. The new function is an inverse function so as to improve the original index gradient threshold function which tends to zero earlier. Gradient threshold function solving the problem of fixed gradient threshold making the edge blurred. But it’s too earlier to be zero, so that it can’t achieve the goal of fully denoising. The New gradient threshold function will reduce the speed of the gradient threshold tends to zero.2. This paper also analyses the improved nonlocal means denoising method based on global image, describes a new one with image structure analysis and edge detection. Nonlocal means method mainly rely on looking for similar image block in the global scope, and calculating contribution weights of every pixel through similarity of image block. But there are some differences in the image block size between smooth areas and structure areas. So this paper uses image edge detection results to adaptively choose the most suitable edge structures and to improve accuracy of similarity measure. Finally, this paper improves the denoising method which combines nonlocal means and sparse representation. So that it can fully use advantages of nonlocal means in dealing with pseudo texture and the ability of sparse representation to tackle the non-smooth area.Through experimental comparison and analysis, the improved algorithm is effective to protect the edge of the image and eliminate the pseudo textures, improving the peak signal to noise ratio.