The Application Of The Partial Differential Equations Diffusion Model In Image De-noising

With the constant development and advance of digital technology, digital picture processing technique widely is used in biometric features recognition, traffic management, industrial and military fields. Images are inevitably interfered by various kinds of factors in the process of acquisition, management, compression, transmission and decompression, thus images be polluted by noise, which degrade image quality. Therefore, image de-noising research has the practical significance. As partial differential equation has been aroused common concern by international community. Meanwhile, benefited form well-understood mathematical theory of partial differential equation, a large number of image processing model based on partial differential equation have sprung up.Firstly, this paper present several kind of classics partial differential equation of image de-noising. According to the disadvantage of P-M nonlinear diffusion model and the self-snake model during the de-noising process, novel image de-noising method based on mixing self-snake model and P-M diffusion model is proposed, these two models use different diffusivity. In order to balance noise diminish and protection of detail feature such as edges and texture, a fidelity term is added into mixed model. Secondly, this paper analyzes the advantages and disadvantages of linear and nonlinear structure tensor in detecting local feature direction of image, Given traditional structure tensor is constructed by tensor product of image gradient, this often leads to don’t effectively detect some special local structure of image, so Hessian matrix is introduced as structure tensor to together to construct a diffusion tensor with nonlinear structure tensor. When constructing diffusivity, we consider edge enhancement and coherence enhancing diffusion.Finally, the simulation experiment results show that the proposed method not only effectively eliminate noise, but also well retain detail feature such as edge and texture, at the same time, processed image don’t change in form, proposed method can improve the visual quality of image.