Improved optimization methods for image registration problems
- 151 Downloads
In this paper, we propose new multilevel optimization methods for minimizing continuously differentiable functions obtained by discretizing models for image registration problems. These multilevel schemes rely on a novel two-step Gauss-Newton method, in which a second step is computed within each iteration by minimizing a quadratic approximation of the objective function over a certain two-dimensional subspace. Numerical results on image registration problems show that the proposed methods can outperform the standard multilevel Gauss-Newton method.
KeywordsImage registration Multilevel strategy Gauss-Newton method Subspace methods
Unable to display preview. Download preview PDF.
This work was partially supported by UK EPSRC (grants EP/K036939/1 and EP/N014499/1), by the Newton Research Collaboration Programme (grant NRCP 1617/6/187) and by the National Council for Scientific and Technological Development (grant CNPq 406269/2016-5).
- 13.Yuan, Y.: A review on subspace methods for nonlinear optimization. In: Proceedings of International Congress of Mathematicians. Seoul (2014)Google Scholar