CNET también está disponible en español.

Ir a español

Don't show this again

The Mandalorian season 2 Apple One launch NASA's 'Greater Pumpkin' Spiders with legs that hear Google's Halloween Doodle game CDC on trick-or-treating risks Charlie Brown's Great Pumpkin

Forensic noise analysis helps to identify tampered photos

A new forensic method to identify tampered photos has been developed. It works by detecting inconsistencies in digital noise.

Hate digital noise? Perhaps it's not so bad after all.

All digital cameras leave some trace of noise or imperfections on a photo when it is created. For most photographers, this noise is not a desirable feature. However, a new method of forensic analysis, developed by a team at the University of Albany, turns noise into something quite amazing.

Professor Siwei Lyu has worked on a technique that measures inconsistencies in a photograph's noise profile by analysing spliced images. For example, the image below of Tiger Woods has one distinct noise pattern created by one particular camera, while the flamingo will have a different noise pattern.

(Credit: Fourandsix)

Analysing this noise data can give a good indication of possible tampering, as indicated in the right-hand-side frame.

There are other solutions available on the market that differ from this model, including encryption algorithms baked in to high-end Canon and Nikon cameras. When a photo taken with one of these cameras is run through image-verification software, it will determine whether a photograph is genuine.

Both of these systems have been cracked over the past few years by teams of programmers and security experts, which leaves the door open for other solutions, like the one outlined by the University of Albany team.

As outlined by photo authentication company Fourandsix on its blog:

This is a particularly challenging problem, because estimating noise is a highly unconstrained problem. By exploiting statistical regularities in natural images, the problem of noise estimation can be more constrained. And, most impressively, this estimate can be performed on relatively small image blocks.

The method is not foolproof, though, as false positives may arise depending on the actual image content and if the forger successfully introduces a uniform noise pattern across the tampered image.

A full outline of the methodology can be found in the team's paper (PDF).