This project was done by Jafar Badour, Hussain Kara Fallah and Almir Mullanurov.
The goal of Single image super-resolution (SISR) is to recover the lost details from a low-resolution (LR) image providing a high-resolution (HR) image.
The main issue we face with such task is that there are multiple HR images that could produce an LR image upon down-scaling, gaussian noise effects or similar effects on the image from physical sensors or the refractive index of the medium of which the light passes.
The challenge that remains after extended research around SISR is how to recover photo-realistic
results with more natural textures and less abnormal extra textures added by the models. The learning
method we will mention in this report and other learning methods that tackle the same problem,
constructs a non linear mapping between an LR image to an HR image H_r = f(L_r, t)
where we have f is the model functional representation that produces a mapping between L_r,
low resolution image and
TBA