[Home] [Abstract/Resume] [Thesis text] [Programs] [Publications]
This page presents my PhD thesis, defended in July 2001 at Ecole Polytechnique Fédérale de Lausanne. Feel free to mail me if you are interested.
Suppose we have two images which we shall call reference and test. These images depict the same or similar object but they are not identical. The task of the image registration is to find homologous (corresponding) points in both images. For us, the output of the image registration is a correspondence function which takes a coordinate of any point in the reference image and returns a coordinate of a~corresponding point in the test image. We call our registration elastic, because the correspondence function can be almost arbitrary.
Most often, the correspondence function is used to warp (deform) the test image in order to align it with the reference image. Other applications include motion analysis, measuring trajectories, distances, speeds, and accelerations, video compression, object tracking, image stabilization, segmentation, and 3D reconstruction. In the biomedical domain, image registration is used to align images prior to comparison, or to quantify the difference between the images. We can compare images of different subjects, taken at different times, or of different modalities. It also helps to create and use atlases, and compensate for inherent geometrical distortion or unwanted motion.
As an example, we register two slices of anatomical MRI images from two different subjects. We can see on the bottom-left image the misalignment of the two images which is corrected by the registration and warping.
|
|
| Reference image | Test image |
|
|
| Before | After |
There is also a movie of the registration process in the QuickTime format.
You can read the abstract in English of
French. The complete text of my thesis is available
in compressed PostScript or
Pdf formats.
The programs I developed are
also available, as well as the slides
of my public presentation.
[Home] [Abstract/Resume] [Thesis text] [Programs] [Publications]