open-discussion > Scaling in DARTEL, diffeomorphic registration
Showing 1-1 of 1 posts
Jan 17, 2017 01:01 PM | Theresa Strauch - Fraunhofer Institute for Solar Energy Systems
Scaling in DARTEL, diffeomorphic registration
Hello,
I use DARTEL, diffeomorphic registration, to register images (non-medical, tif format) with each other. The results show the expected vector direction, but the scaling is wrong - more than 2 orders (factor 500-1000) beneath the ground-truth in pixels.
The input images are aligned and normalized gray scale images with dimension 1024*1024*1*1. It seems that the scale of the flow is connected to the image intensity because the images I used as input are brighter in certain areas, and in those areas the flow is larger. But as the algorithm converges I suppose the problem is not inside the iterations.
In the attached file you can find the specifics about how I called the function from MATLAB but I think the question is general (I used many different parameter sets, different models and time steps, but the problem appeared independently of the setup).
I would be grateful for any help -
Theresa
I use DARTEL, diffeomorphic registration, to register images (non-medical, tif format) with each other. The results show the expected vector direction, but the scaling is wrong - more than 2 orders (factor 500-1000) beneath the ground-truth in pixels.
The input images are aligned and normalized gray scale images with dimension 1024*1024*1*1. It seems that the scale of the flow is connected to the image intensity because the images I used as input are brighter in certain areas, and in those areas the flow is larger. But as the algorithm converges I suppose the problem is not inside the iterations.
In the attached file you can find the specifics about how I called the function from MATLAB but I think the question is general (I used many different parameter sets, different models and time steps, but the problem appeared independently of the setup).
I would be grateful for any help -
Theresa