Three dimensional locally adaptive warping for volumetric registration

Abstract

Locally-adaptive, sub-voxel registration of two volumetric data sets is performed by generating match points and then, using the match points, generating a locally-adaptive image-to-warp transformation. Generation of the match points is accelerated by generating a volume hierarchy and then identifying a set of 3D points in the first volume for designation as interesting points. Interesting points in the first volume are then matched with their corresponding points in the second volume. After performing match point generation, points on a cubic grid in the first volume are matched to the corresponding match points in the second volume by resampling the match points on the cubic grid in the first volume. After a grid of match points has been identified, the displacement (dx, dy, dz) that should be added to each cubic grid point in the first volume to identify the coordinates of the corresponding point in the second volume is determined. Using the displacements in the match points, a warping transformation function that specifies the displacement for each point (x, y, z) in the first volume is determined. The transformation function maps the matched points in the first volume to corresponding points in the second volume by determining the amount of displacement in the x, y and z directions.