Image Denoising using Stochastic Differential Equations

We address the problem of image denoising using a Stochastic Differential Equation approach. We consider a diffusion process which converges to a Gibbs measure. The Hamiltonian of the Gibbs measure embeds an interaction term, providing smoothing properties, and a data term. We study two discrete approximations of the Langevin dynamics associated with this diffusion process: the Euler and the Explicit Strong Taylor approximations. We compare the convergence speed of the associated algorithms and the Metropolis-Hasting algorithm. Results are shown on synthetic and real data. They show that the proposed approach provides better results when considering a small number of iterations.