Anisotropic Diffusion Model With Detail-preserving For Image Denoising

Image is an easy, compact, and widespread way to represent the physical world, so the physical world could also very well be defined as an image society. In order to get high quality image, digital image processing and analysis play an important role. As one of the oldest concerns, image denoising is still a necessary pre-procession step for many applications.The development of image denoising and computer vision is closely linked. With the rapid changes of science and technology, the traditional image denoising algorithms cannot satisfy people’s growing demand for high quality image. In recent days, many complicated mathematical tools have received extensive attention. As one of the most important part of mathematical analysis, partial differential equation(PDE) has been widely used in the field of image processing and analysis.Traditional image filtering algorithms can smooth the noise, but also can cause the loss or fuzzy of the important detailed information of images such as more texture, weak boundaries and so on. The denoising effect is always not ideal. The methods based on nonlinear diffusion equation can suppress the noise effectively, which make then be widely used by researchers. Anisotropic diffusion model for image denoising is a hot research field, in which the diffusion coefficient plays an important role. The coefficient usually depends on some local characteristics of image. It can denoise the flat part of image and enhance the edge portion adaptively.Both the classical P-M model and the total variation(TV) model are based on image local characteristics(gradient) to realize the image denoising process. The gradient values are used to construct a diffusion function or as the measure of image smoothing. But these models are easy to result in the missing or fuzzy of the fine features without considering comprehensively the image detail features.In order to retain more meaningful information or other fine features of low contrast, this paper proposes a new PDE for image denoising that can preserve details well. The model is based on the P-M model and curve evolution theory. This PDE is made up of diffusion term and an image adaptive fidelity term. The diffusion term employs the normalized variance to improve the diffusion efficient, which makes the model can remove the noise better. The image adaptive fidelity term can realize the remove of noise and the sharping of image edges selectively. Experimental results show that the proposed model not only can remove the noise effectively, but also can reserve more important detailed feature than other models.