Deep Learning For Image 'Enhancement' | Super-Resolution Series #2 - Concept Review

Super Resolution (SR) is a really cool application of machine learning that is used to create high-resolution images from low-resolution inputs. In this video, I provide a comprehensive overview of SRCNN, a well-known method of implementing SR that utilizes convolutional neural networks following the work of Dong Et al (2015). I also cover an alternative quality metric, following the work of Wang Et al. (2004), known as Structural Similarity Index (SSIM) that can be used to better approximate the Human Vision System (HVS) when measuring reconstructed image quality. By no means am I an expert on any of the concepts in this video, there are most likely some misconceptions I made during my research process, so if you notice something off please let me know in the comments!

The next video will cover the implementation process, so if you’re interested in seeing that feel free to check my community tab for updates!

I made all the animations in this video using ManimCE, it’s a really great resource for making fun animations, so if you’re into that kind of stuff, I’d HIGHLY recommend checking out their website and playing around with the library…

SSIM Demo: https://github.com/brettmorrisonAO5AN...
Manim Code: https://github.com/brettmorrisonAO5AN...
ManimCE Website: https://docs.manim.community/en/stabl...

Sources:

(SRCNN Paper): https://arxiv.org/pdf/1501.00092.pdf
(SR General Overview): https://ieeexplore.ieee.org/stamp/sta...
(Sparse Coding): http://ufldl.stanford.edu/tutorial/un...
(SSIM Paper): https://ece.uwaterloo.ca/~z70wang/pub...
(Sparse Coding For SR): https://www.columbia.edu/~jw2966/pape...
(Data Processing Inequality): https://en.wikipedia.org/wiki/Data_pr...
(PSNR): https://en.wikipedia.org/wiki/Peak_si...
(SR Overview):    • How Super Resolution Works  
(DeepVis ToolBox):    • Deep Visualization Toolbox  
(SSIM Example):    • Structural similarity and image proce...  

Keywords: machine learning, deep learning, convolutional neural network, image processing, computer vision, python, manim, super-resolution, structural similarity index, sparse coding