How is blurring added to photos and videos and how can be removed?

607 views

How is blurring added to photos and videos and how can be removed?

In: Technology

Natural blurring in photos and videos is from the focal point and depth of field. This cannot be removed. Blurring by software can be altered but will be different in case by case scenarios.

Artificial blurring in photos (or videos; for our purposes those are just a series of photos) is similar to smudging the wet paint in a painting:

For every point in the area that is to be obfuscated, the colour is replaced by an average of the colours of all the points in a certain radius. If instead of calculating this average for each point you calculate them for a group of points at a time, you get a pixelated look instead of a blurred one.

As for reversing that effect:

This really depends on the radius used to calculate the blurring: A higher radius will result in a more blurred image and will effectively erase more information.

With well-known possible images underneath the blurring it is somewhat feasible to recover some information: There have been numerous examples of people uploading photos/ screenshots with text in them that they didn’t want people to see only to use a low radius – those texts could even be recovered by hand years ago, so I suspect that there is now some machine-learning tool out there that can read those automatically. Nowadays you can actually do quite a bit with machine-learning when your input is somewhat consistently formed – there is even DeepCreamPy, which automatically deobfuscates hentai pictures (no, it doesn’t work on real porn).

But for completely random input? There is no real way of knowing what was behind the blurred area. I suspect that in the coming years, people might create a neural network that could produce somewhat realistic results for low-detail areas (say, a blurred facade on Google maps) if you throw enough training data at it, but for high-detail stuff like faces, there really is no way of reconstructing them if they are blurred well enough.