Majorize–minimize linesearch for inversion methods involving barrier function optimization

This paper focuses on the issue of stepsize determination (linesearch) in iterative descent algorithms applied to the minimization of a criterion containing a barrier function associated with linear constraints. Such an issue arises in inversion methods involving the minimization of a penalized criterion where the barrier function comes either from the data fidelity term or from the regularizing functional. In order to circumvent the inefficiency of general-purpose linesearch strategies in the case of barrier functions, we propose to adopt a majorization–minimization scheme by deriving a new form of a majorant function well suited to approximate a criterion containing barrier terms. We also establish the convergence of classical descent algorithms when this linesearch strategy is employed. Its efficiency is illustrated by means of numerical examples of signal and image restoration. (paper)