Deblur (The Tool Jam 2)

When an image is blurred, a lot of information still remains. This tool attempts to reverse Gaussian and Box-Filter blurs using a user-provided estimate of the blurring radius. It can also create interesting graphical effects, or blur and deblur normal images for fun.

Modes

The tool has two modes, Blur & Deblur and Deblur. You can switch between them using the dropdown menu at the top left. 

Blur & Deblur mode takes a normal image, blurs it (using the settings at the bottom left), and then deblurs it (using the settings in the center and bottom right). The images from left to right are: the original image, the blurred image, and the deblurred image. The deblurring process does not use the original unblurred image at all, just the blurred one. You can upload your own images using the dropdown selector labeled "Original Image".

Deblur mode takes a blurred image, and tries to deblur it. The four images it shows are the image it's trying to deblur (top-left), the current best guess (top-right), the best guess after it's been blurred (bottom-left), and the current error (bottom-right). You can import your own blurred images using the dropdown selector labeled "Blurred Image".

Technical Details

Deblur uses an iterative "guess and check" approach that's designed to converge towards an optimal unblurred image. It basically makes a guess, blurs it, compares that blurred guess to the target blurred image, randomly adjusts the guess in areas that differ, and repeats. 

You'll notice that it doesn't usually find a good result on its first pass though. For best results, try nudging the blur radius up and down repeatedly until the image becomes clearer (you can use the left and right arrow keys to do this).

Notice on Ethics

Please don't attempt to invade others' privacy with this tool.

This tool is intended to be educational, entertaining, and maybe help generate some glitch art. In practice, I've only been able to achieve great results if I know precisely how the blur was performed (in other words, if I blurred the image myself), so I think it's unlikely this could be used in many real-world applications anyways.

There are more variables than just radius when it comes to blurring, and unless you have them all correct, you hit a wall pretty quickly with this technique. (I've also purposely not included some features like partial-image deblurring, since that's pretty much only useful for revealing things that were meant to be hidden).

That said, blurring is not an effective way to conceal information in photos. A determined attacker can uncover more than you'd expect.

Made with python, pygame, and pygame_gui
Source code can be found here.