3D Mesh Alignments Using Multi-resolution Non-rigid Image Registration Techniques

Montag, 28. Juni 2021

This project is my master thesis which illustrates a point-set registration problem which is also viewed as a 3D binary images registration problem.
Point-set registration and image registration are typical and fundamental problem in image processing.
Point-set registration aligns two sets of point clouds or meshes.
Image registration is a process of comparing and fusing images into one image by transforming the data into a single coordinate system.
It is an important task which could be applied to a wide range of applications such as in computer vision,
automatic target recognition. In biomedical image analysis, image registration is used for change detection or tumor monitoring.
By converting the point-set registration problem to image registration problem, spatial transformation is required without point-to-point correspondence.
Rigid transformation follows by non-rigid transformation are applied.
Multi-resolution scheme is also used to vary the parameters in different resolution levels.
Experiments are conducted on a public spine dataset. An overall average dice is 0.92.