I am working on the project where my goal is to create SR image from 25 LR images that are precisely shifted by 0.2px (both in horizontal and vertical direction ). I also know that the captured object is a back and white grid ( something like chessboard) I am looking for some precise SR method that will take advantage of the precise shift of the images. The emphasis is on precision not on the speed of the method. I have read a lot of the papers on that topic and found out that there is a lot of methods and it is difficult for me to decide which method suits my task.

If you have experience with SR please suggest some methods that find good for this task.

  • $\begingroup$ This is for a class? The class didn't talk about specific methods before this assignment? $\endgroup$
    – endolith
    Nov 3 '17 at 18:24

You'll need to describe the density ("kernel") with which light from different angles contributes to each (physical) pixel's amplitude. Expect something rather gaussian.

Now, shift that kernel by 0.2 pixel width, repeat. You'll find a Max-Likelihod estimator for the high-density picture simply by finding a system of equations that describe how much the values of the 5 pixel positions whose densities overlap that position contribute.

If you want a very intuitive first approach, look for the kdeplot tool of the python seaborn library. The website has a nice gallery.


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.