In this paper, we introduced two algorithms (SFP and MG) for solving the curvature-based denoising model [56], which is high order and capable of effective noise removal. The resulting EL equation is of fourth order, anisotropic, and highly nonlinear, so conventional algorithms struggle to find the solution quickly and efficiently. In contrast, our MG algorithm is shown to be fast and robust up to some degree to changes in the noise level and parameters.
We explained that a fixed point method using the Vogel and Oman idea [51] is unstable and simply does not work for this curvature-based formulation. We then showed how to stabilize this fixed point method and developed a stabilized fixed point method, giving evidence through local Fourier analysis of its smoothing properties. Based on this, we developed a fast nonlinear MG method. Finally, a generalization of our algorithms to similar problems was discussed. A recent generalization to color images was done in [12].