On Digital Image Denoising

Digital image processing is a new field derived from mathematical technology and computer science. Image denoising is one of the widely discussed topics in this field. Effective denoising is critical for digital image information pre-processing. During two decades' development, variety of denosing models were proposed, such as Gaussian filtering, median filtering, wavelet transformation and DCT filtering, anisotropic diffusion model, minimum total variation model, bilateral filtering, non-local means filtering e.t.In this paper, we discussed the original of anisotropic diffusion model, its discrete numerical scheme and relation with other models. We also introduced bilateral filtering and non-local means filtering, and all these models' advantage and disadvantage were studied. The last part of this paper are our related works. We substituted Gaussian convolution with bilateral filtering in Catt model, bilateral filtering is an anisotropic filtering, this technique made edge detection more precisely. We also introduced modified force term to accelerate the iterative speed, at the same time get good denoising performance. Another work of this paper is extending bilateral filtering to non-local trilateral filtering. Trilateral filtering can make full use of the high degree of redundancy of natural images, and get impressed results in low noisy and natural image denoising. At last we proposed a new non-local algorithm combining NL-means and trilateral filtering based on image decomposition. We introduced a texture detector to decompose the image into two parts. In the texture dominant part, we use NL-means filtering, in the smoother part we apply trilateral filtering. Our algorithm reduces the computational cost, with good denoising performance.