CODEWITHSUNDEEP
pixelimaging
Swimming
Swimming

Reputation: 11

How to do pixel super-resolution?

I am currently exploring pixel super-resolution algorithms and am unsure where to begin, so I have attempted to replicate the work of others. The paper I am trying to replicate is titled "High-resolution image reconstruction from a sequence of rotated and translated frames and its application to an infrared imaging system".

The paper describes the low-resolution (LR) images as follows: Y = HX + n, where Y is the LR image, X is the super-resolution (SR) image, H is a weight matrix that describes image rotation, translation, and blurring, and n is a noise term. In the algorithm described in the paper, X is the unknown variable and can be computed by iteration. In the final result, (Y-HX) reaches the minimum value.

To replicate the algorithm, I have used Matlab. First, I loaded a 256 x 256 image and then downsampled the image in four directions to obtain four LR-images with a size of 128 x 128. However, I am unsure how to find the H matrix if assuming the shift distance is unknown.

image = imread('cameraman.tif');
lr_image1 = image(1:2:end,1:2:end);
lr_image2 = image(2:2:end,1:2:end);
lr_image3 = image(1:2:end,2:2:end);
lr_image4 = image(2:2:end,2:2:end);

Really appreciate if anyone could help me out.

Upvotes: 1

Views: 171

Answers (1)

Philip Day
Philip Day

Reputation: 69

LET ME BE CLEAR - I DON'T KNOW THE ANSWER

With that out of the way, I can across the thread because I'm interested in the subject, but coming from an even lower starting position than OP.

The paper which you have cited mentions: "The key to the success of the algorithm is having an accurate observation model. This includes the image registration parameters and the system PSF."

They used an iterative gradient based technique to estimate those parameters, which might help you in your challenge?

Sorry not to be more helpful

Upvotes: -1

Related Questions