Two Fast Nonoverlapping Domain Decomposition Methods for the Dual ROF ModelPP

We propose two nonoverlapping domain decomposition methods (DDMs) for the dual Rudin-Osher-Fatemi (ROF) model. A primal DDM is proposed, which resembles the classical Schur complement method. It achieves $O(1/n^2)$ convergence, the best rate among the existing DDMs. A primal-dual DDM based on the method of Lagrange multipliers on the subdomain interfaces is also considered. Local problems can be solved in linear convergence rate, while the standard algorithms for the ROF model are $O(1/n^2)$ convergent.

This is poster number 32 in Poster Session

Jongho Park (KAIST)
Chang-Ock Lee (KAIST)
domain decomposition methods, dual total variation, image reconstruction, nonlinear optimization