Nonlinear Inverse Scale Space Methods for Total Variation Blind Deconvolution

In this paper we propose a blind deconvolution algorithm based on the total variation regularization formulated as a nonlinear inverse scale space method that allows an efficient recovery of edges and textures of blurry and noisy images. The proposed explicit scheme gives the restored image solution by evolving in time the zero signal and an estimated kernel until a stopping criterion is satisfied. Numerical results indicate that our scheme is robust and converges quickly to the solution of the model for images convolved with either a Gaussian-like experimental point spread function or Gaussian blur and contaminated with Gaussian white noise.