Anisotropic Diffusion Model For Image Denoising

With the rapid development of computer science, demands for digital imageprocessing are also increasingly expanding.Common image denoising algorithmincludes a variety of adaptive median filtering algorithm, wavelet thresholdingalgorithm, based on partial differential equations, total variation minimizationalgorithms, non-local means filtering algorithm, and so on. Denoising is a pretreatmentprocess of image processing,it is an important branch of image processing. PDE imagedenoising method is a hot research field, attracted widespread attention at domestic andabroad scholars. In many methods, the anisotropic diffusion model has become apopular method of PDE for image denoising.Entropy has important applications in probability theory, information theory,physics, graphics and other areas, and has made many important research results. Which,Tsallis entropy has been widely used in many areas of image processing, such as imagethresholding segmentation, image matching. Tsallis entropy is the promotion ofShannon entropy, and has the following properties:1)the isolation of noise impact onthe local Tsallis entropy is very small.2) local tsallis entropy is higher at image details.This dissertation mainly does the following research:The diffusion coefficient determines the extent of the diffusion model inanisotropic diffusion model, which is a function about the image gradients, at gradientbigger place, the diffusion coefficient is small, the gradient of the smaller places, thediffusion coefficient is big. Therefore, the model can smooth the image while the imageedge retention. However, this model also has some disadvantages, only use of gradientinformation of the image, the image texture and the weak details are not wellmaintained. To address this problem, this paper combined with image Tsallis entropyentropy which is different at the edge of image and the smooth area of the image, weproposed anisotropic diffusion modelcombine with Tsallis entropy, the diffusioncoefficient of the new model is not only dependent on the image gradient, and use of theimage local Tsallis entropy, the experiment showed that: the new model for the denoisedimage can better retain the detail and weak edges of image, and has a good denoisingperformance.