I have already worked on medical image registration. the first and the most important challenge in rigid body image registration is finding a good similarity measure. This measure should be maximized in order to achieve best alignment between two images. mutual information is one of the est examples. the second challenge is the optimization method by which we want to find transformation function parameters. Most of the optimization methods converge to local maximums which is not desired.

here you can find good medical image data;

http://www.insight-journal.org/rire/download.php

good luck

I am working on medical image registration, but in different aspects; 3D registration,

In my work, I have two kind of 3D registration; using MRI images, and using depth map images. Let's begin with MRI which is a bit easier than the other one

Regarding the MRI, it is required to perform necessary segmentation to remove parts which you don't need, after that, since layers are sorted in correct orders, you can assign a sufficient value as the third dimension to the layers and construct the 3D model step by step.

The second case in difficult. It seems to solve a 3D puzzle since you have to find the overlaps between layers and align them correctly, that mean you have to deal with 4 parameters, x,y,z for movement and rotation. If the scanned model has non-rigid properties, thing will be worse, while you need to find skew and scale!

Additionally if the scanned data are far form each other, you may need to perform coarse registration step in advance to fine registration.

You can find good details here:

https://www.researchgate.net/publication/267748684_3D_Reconstruction_of_Body_Parts_Using_RGB-D_Sensors_Challenges_from_a_Biomedical_Perspective

Conference Paper 3D Reconstruction of Body Parts Using RGB-D Sensors: Challen...

Conference Paper Benchmarking of Image Registration Methods for Differently S...

Framework - https://borda.github.io/BIRL/

We have organised Automatic Non-rigid Histological Image Registration (ANHIR) challenge which was a part of the IEEE ISBI conference 2019.

https://anhir.grand-challenge.org