Image registration is a powerful tool in various tomographic applications. Our main focus is on microCT
applications in which samples/animals can be scanned multiple times under different conditions or at different
time points. For this purpose, a registration tool capable of handling fairly large volumes has been developed,
using a novel pseudo-3D method to achieve fast and interactive registration with simultaneous 3D visualization.
To reduce computation complexity in 3D registration, we decompose it into several 2D registrations, which are
applied to the orthogonal views (transaxial, sagittal and coronal) sequentially and iteratively. After registration
in each view, the next view is retrieved with the new transformation matrix for registration. This reduces the
computation complexity significantly. For rigid transform, we only need to search for 3 parameters (2 shifts, 1
rotation) in each of the 3 orthogonal views instead of 6 (3 shifts, 3 rotations) for full 3D volume. In addition, the
amount of voxels involved is also significantly reduced.
For the proposed pseudo-3D method, image-based registration is employed, with Sum of Square Difference
(SSD) as the similarity measure. The searching engine is Powell's conjugate direction method. In this paper,
only rigid transform is used. However, it can be extended to affine transform by adding scaling and possibly
shearing to the transform model. We have noticed that more information can be used in the 2D registration if
Maximum Intensity Projections (MIP) or Parallel Projections (PP) is used instead of the orthogonal views. Also,
other similarity measures, such as covariance or mutual information, can be easily incorporated.
The initial evaluation on microCT data shows very promising results. Two application examples are shown:
dental samples before and after treatment and structural changes in materials before and after compression.
Evaluation on registration accuracy between pseudo-3D method and true 3D method has been performed.