Michael J. Allison1, Jeffrey A. Fessler1
1Electrical Engineering and Computer Science, University of Michigan, Ann Arbor, MI, United States
Existing regularized field map estimators are highly robust, but require the minimization of a non-convex cost function. The current fastest minimization method, an optimization transfer approach with separable quadratic surrogates, requires thousands of iterations to converge. We propose a novel optimization transfer method which uses Huber's algorithm for quadratic surrogates to solve the non-convex problem. By framing the problem in this way, we are able to exploit the sparse banded structure of typical finite differencing matrices. We evaluated our algorithm on a brain image dataset finding that it converged in one hundredth of the time.