Mosaic vs Gaussian Blur: Which Actually Protects Identities (and Which One AI Can Undo)

You pixelated a face in a screenshot, posted it, and someone un-pixelated it well enough to recognize the person. This isn't a horror story — it's a documented technique. Face blurring feels like a permanent lock, but it's a filter applied to known pixels, and some filters can be reversed. Choosing the right blur is the difference between "I blurred it" and "it's blurred."

Why Some Blurs Come Back

A heavy face blur destroys most of the information in a region — that's the point. The problem is the light blur: a subtle gaussian or a thin mosaic keeps enough structure that a model trained to reverse it can reconstruct a plausible face. Research has shown faces can be recovered from blurred images with the right algorithm. The common mistake is applying "a little blur" to be polite, which is about as private as a scribble over a name you can still read.

How to Blur So It Stays Blurred

Strong, aggressive removal beats gentle filtering. Use a heavy mosaic or a solid mask rather than a soft blur — the fewer original pixels survive, the less a reconstruction has to work with. The counter-intuitive part: a completely opaque shape covering the face — not a translucent blur — is the honest version of hiding someone. If the goal is anonymity, removal beats blurring entirely: the object remover deletes a face from the scene rather than smudging it, and the background remover helps when the whole context needs to go.

The Policy Question Nobody Sets

Before you blur, decide what "blurred enough" means for your use case. A news still where the face is incidental and the person isn't identifiable — fine. A courtroom photo where identity is the whole point — nothing less than full removal or masking is defensible. We compared pixelation, masking, and blurring in our guide to face hiding methods. Choose the level of destruction that matches the risk — not the level that looks tidy in a thumbnail.