Neural Enhance

A not-quite-CSI image enhancer. FREE | BIT.LY/NEURALENHANCE

We already know how to imitate a Hollywood technodrama in Linux using Ubuntu’s “Hollywood package”, but what if some magic from secret CSI labs could become real and we could, for example, enhance still shots from CCTV cameras in order to read numbers on car plates? Well, we’re getting a little closer to it thanks to Neural Enhance.

There seems to be a growing trend of open-source algorithms that use neural networks to enhance pictures. We surfed GitHub and found many similar projects, but Neural Enhance is the most noteworthy. It’s also quite simple to set up.

The idea behind Neural Enhance is trivial: although a lowresolution image can’t be enriched with more real details, we can reconstruct and suggest such details using a data model, which is based on other images. As a result, it becomes possible to upscale an image with greater quality than with convenient methods (such as ‘Image > Resize in GIMP’). The quality of the output is very dependent on such a model. But Neural Enhance is quite friendly to beginners and offers a viable test model by default. Installing the whole thing can be complicated for nontech-savvy Linux users, but luckily the author offers a quicker way by using Docker. If you want to discover the possibilities of the algorithm manually, just install Docker for your distribution (distro) and then run the Neural Enhance image:

$ docker run alexjc/neural-enhance /bin/ bash

A more practical way is to create an alias that automatically pulls the proper image and passes the command:

$ alias enhance=’function ne() { docker run --rm -v “$(pwd)/`dirname ${@:$#}`”:/ne/ input -it alexjc/neural-enhance ${@:1:-1} “input/`basename ${@:$#}`”; }; ne’ Then upscale your image with: $ enhance --zoom=2 --model=small images/ example.jpg

Use a different value for the the scale factor. --zoom option to adjust