Image Matching Using Convex Optimization

Faculty: Camillo J. Taylor

Project Homepage:

Arvind Bhusnurmath and CJ Taylor have been working on approaches to recasting many classic image matching problems including, stereopsis, motion estimation, image registration and 3D volumetric matching as convex optimization problems that can be solved effectively using the Interior Point method. More specifically they proceed by constructing piecewise linear convex approximations of the original image matching functions and then reformulate the matching problems as linear programs. Importantly, in each case they are able to exploit the structure of the resulting linear program to develop efficient algorithms which allow them to solve optimization problems involving hundreds of thousands of variables more efficiently than standard codes like TOMLAB and MOSEK.