A controllability approach to shape identification (Q953892): Difference between revisions

Automatic registration of image pairs pervades artificial vision, starting with stereo-vision, and optical flow in video sequences. In biomedical imaging, automatic matching for 3D images of soft organs across different subjects is an important step, and quantifying dissimilarities between organ shapes impacts clinical diagnosis. Nonlinear image registration relies on minimizing cost functionals combining two terms, smoothness and disparity. A powerful mathematical approach, linked to geodesics in infinite dimensional Lie groups of diffeomorphisms of \(\mathbb R^ 3\), was successfully explored for soft shape matching, i.e. [\textit{P.~Dupuis, U.~Grenander} and \textit{M.~I.~Miller}, Q. Appl. Math. 56, No. 3, 587--600 (1998; Zbl 0949.49002)]. In this paper, the authors formulate the matching problem as an approximate controllability problem involving a cost functional whose gradient is computed by using an adjoint equation based methodology. Also, they discuss the time discrete version of the image matching problem.