Research On Diffusion Model And Image Denoising Based On GVC

A large number of images need to be processed these days. In many fields, one wishes to interpret given images, which are often corrupted by noise, such as medical images, remote sensing images, archaeological images and so on. Therefore, image denoising has an important practical significance.During the last two decades, partial differential equation (PDE) based denoising methods have been widely applied and studied. Second order P-M model and fourth order Y-K model are two classical PDE-based denoising algorithms. However, P-M model can bring staircases effect in the resulting images. These staircases can be falsely detected as edges in the successive image processing. Although Y-K model can effectively eliminate the staircase effect because this fourth-order PDE favors a piecewise harmonic image, this model is an isotropic filter which will causes the loss of edge and texture information.In light of the above defects, novel second order and fourth order denoising methods are proposed in this paper. Both models are based on gradient vector convolution (GVC). Since GVC have great robust on Gaussian noisy and can be implemented in real time, these two new methods have better performance of denoising effect, edge preserving and calculating speed.In this paper, we first introduce the GVC model into the anisotropic P-M equation. This model improves the numerical stability of inverse diffusion term so that it can better sharpening edges and smoothing the regions that are relatively flat.Second, this paper proposes a new fourth order diffusion model based on GVC. The proposed fourth order model is an anisotropic filter while the original Y-K model is an isotropic diffusion model, so the GVC-based Y-K model can obviously improve the ability of edge and texture preserving besides further improvement of denoising.