1
$\begingroup$

I have two images of two objects that are suppose to be exactly the same. I use one for reference and one as my model. I want to find an algorithm that will help me find on which areas these images vary? and return a binary map (the size of the image), that indicates of the variation. Its an easy task when we don't take into acount noise and small shifts, but in real life it does not work like that.

What is considered state of the art to tackle that problem in a way that would fit such factors? Thanks!

$\endgroup$

2 Answers 2

1
$\begingroup$

On top of other suggestions, feature registration may be appropriate if the images are difficult to register. Detect features using an algorithm like SIFT, and then use RANSAC to find correspondences between the images.

$\endgroup$
0
$\begingroup$

You'll have to experiment with methods. The exact approach is likely to depend on the specifics of exactly what kinds of variations occur and you want to ignore. Start with https://en.wikipedia.org/wiki/Image_registration and https://en.wikipedia.org/wiki/Image_differencing and SSIM and MSE, and get some experience with in what ways they do or don't need your needs, and that will help you formulate a more precise question.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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