Inclusion and estimates for the jumps of minimizers in variational denoising (Q6623967)

The authors introduced a general approach for studying jump infusion and control in variational denoising problems in this work. They specified conditions at which a jump inclusion occurs and derived an estimate of the magnitude of the jump. The differentiability assumption on the regularizer that was used can be seen as a relaxation of double Lipschitz comparability for convex regularizers. A discussion of general regularizers that satisfy the assumptions of the main result including certain regularizers for vector valued images is given. Conditions that guarantee the local boundedness of the solution of the variational problem are also discussed. Furthermore, the authors addressed adjustments needed for extending the proof techniques to the case of inf-convolution based regularizers. Numerical results for the Rudin-Osher-Fatemi problem were given and the versatility of the approach was discussed.