Figure 3

Schematic overview of a single registration run in 2D, which is in principle similar to the real registration in 3D. Original binary segmented \(\upmu\)CT image and mask are combined into masked image (a). This image is then resampled to the same size and rescaled to match the voxel spacing of the reference. Image after similarity transform using the reference image and undeformed B-spline control point grid with grid spacing g (b). Only grid nodes in the vicinity of the image domain are shown. Image after B-spline transformation with deformed control point grid (c). Transformed original image using the transformation T gained from the two registration steps (d).