Intensity based image registration

Hello,
I have three questions about intensity-based registration:

• how does intensity-based registration work? (detailed explanation /sources to read)
• what is the difference with the methods based on feature detection?
• are there metrics to be calculated to evaluate the quality of registration?

Thank you in advance.

Answer 1

[Ryan Comeau]

Hello,

I am going to provide some commentary, but not mathematics necessarily.

1. Intensity based registration works by minimizing the error in intensities. Imaging you are looking at two waves in an ocean. If these waves were perfectly superimposed, upon their subtraction, their result would be zero. If they are misaligned, their will be some residual upon subtraction. The algorithm will essentially move the image until it superimposes the two images as best as possible.
2. I've never used a featured based method, but if i had to guess, it would use Haar like features on both images and match these Haar like features as best as possible. https://en.wikipedia.org/wiki/Haar-like_feature
3. https://www.mathworks.com/help/images/ref/imregister.html#btcaexy-3. This intensity based registration has a name-value pair which will display the optimization information. This should help you gain some intuition of your mathematical optimization. If you want a human visible idea of your succes or not, just use the

 imshowpair(fixed, movingRegistered,'Scaling','joint')
 

Hope this helps

RC

[D_user]

Ryan Comeau, thank you for your answer.