A computational framework for image-based constrained registration (Q1025865): Difference between revisions

The authors develop a framework for local constrained image registration and apply it to rigidity and volume constraints. It is worth to emphasize that there are a few features that distinguish the method presented in this paper compared to others. The features are as follows: \begin{itemize}\item[1.] The proposed method uses constraints rather than a penalty and in view of this fact the constraints are actually fulfilled. Additionally, using constrained optimization does not require choosing a penalty parameter. \item[2.] The Lagrangian approach saves the need to track the domain of the constraints but adds the cost of computing the determinant of the Jocobian. \item[3.] They use sequential quadratic programming for solving the constrained optimization problem. \item[4.] They examine rigidity and volume constraints. They present that the increase of computations stayed moderate with approximately 15\% compared to a unconstrained approach. At the end, in the numerical experiments they show that adding constraints can increase the reliability of non parametric registration and reduce the inherent non-uniqueness of the problems. As the results they advocate that adding constraints will became common practice in image registrations.\end{itemize}