2
$\begingroup$

I am trying to register two images of daily objects from different viewpoints (5 to 20 degrees angle). I have been reading about image registration and I feel like the best way to go is using Feature-Based methods for my purpose. But I still have some doubts about that.

  1. What is your method recommendation for images like below? Can rigid transformation methods be an option for me?
  2. What is your error measurement recommendation? I use 'root mean square error' currently. And my experiments show that Feature-Based better than Intensity Based methods on these kind of images (At least better than matlab's auto registration).
  3. What is the effect of registration process on the entropy of an image? I feel like registration should decrease the entropy.

Reference Image enter image description here

Target Image enter image description here

Thanks a lot

$\endgroup$
1
$\begingroup$

Take a look to the ImageJ plugin called TurboReg algorithm (or StackReg). It should be really effective for such images.

However, in your example, the two pictures seem to have been taken from different angles (right left), but also different height. So two angles are different, and then you have to find an affine transformation.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.