Investigation of Intensity Correction in the Context of Image Registration

Abstract

An image registration algorithm with intensity correction was developed. A particular goal was to apply intensity correction instead of using multimodal similarity measures.
The algorithm utilises common Levenberg-Marquardt optimisation. The author has chosen two dimensional affine and one dimensional B-Spline model as spatial transformation, as well as intensity correction models specific to CT images. They are global non-linear mapping and smooth local affine correction. The algorithm was tested experimentally using a wide class of simulated images and a limited class of medical images.
Affine registration works properly even for deformations which exceed typical deformation encountered in medical practice. B-Spline registration works properly for small deformations and requires further development to increase capture range.
The idea of separating intensity correction mapping from similarity measure is shown to have advantages. Choosing intensity correction model can make the registration algorithm specific to the image class of interest.