Improved Tensor Diffusion Model for Image Restoration

The local gradient features of image can not be well measured due to the exist of noise, which affects the filtering effect of nonlinear diffusion method. Using two low-level local features of phase congruency and gradient, we propose an improved anisotropic tensor diffusion model. Firstly, a scatter matrix is designed by projection on maximum direction of phase congruency. Secondly, a tensor diffusion model is constructed. Further, a finite difference scheme and the algorithm procedure are listed. Finally, numerical experiments illustrate the computational efficiency of proposed model, objective assessment, and visual effect.