[Insight-users] 1D deformable registration

[Jochen Bertels]

Hi,

I'm trying to non-rigidly register two one-dimensional signals.

My first try was to adapt the DeformableRegistration4 example, by just 
changing the dimension.
This wouldn't compile because the GradientRecursiveGaussianImageFilter, 
which is used in every
imageToImageMetric to calculate the gradient of the metric, can't cope 
with 1D images.

This made me try to mimic the 1D case by using a 2D image with the size 
in one dimension set tot
1. This results in an error saying that the 
GradientRecursiveGaussianImageFilter requires at least 4
pixels in each dimension.

So the only options left, as far as i can see, are
1) use an optimizer that doesn't use gradient information.
2) write my own metric, calculating the gradient another way than with 
GradientRecursiveGaussianImageFilter

I have tried the first option by adapting DeformableRegistration2. This 
works but is rather slow (12 minutes
for a 2638x1 image).

Before trying the second option, i would like to ask if there are other 
options that i haven't thought of.
I find it rather hard to believe i am the first person facing this 
problem, although i haven't found any other
postings on this topic.

Thanks in advance,

Jochen