Using Noise as Camera Fingerprints for Detecting Image Manipulation


A recent photographic analysis technique developed by Professor Siwei Lyu and his team at the University at Albany – SUNY could lead to better forensic analysis of altered images. The technique takes advantage of the fact that, when splicing two images together, each will bring with it the specific noise pattern of the camera it was shot with.

So, when analyzing the obviously fake image at the top, the flamingo Tiger Woods is using in lieu of his golf club shows up as having a different noise pattern than the rest of the image.

Here’s another example of a spliced image and its noise analysis data:


The system isn’t perfect. For example, extensively smooth or textured regions in an image can impact noise detection, leading to false positives. A skilled image manipulator could also, knowing of this technique, introduce a smooth noise pattern.

But even so, this adds another tool to a growing manipulation detection toolbox for forensic analysis. So forgers beware, you can no longer pass off your “Tiger Woods Using a Flamingo as a Golf Club” photo as the real deal.

For all of the technical minutia and more examples, you can read the entire research paper here.

(via Fourandsix)

  • Igor Ken

    seems legit to me…

  • Chris Newhall

    What if I took a high ISO picture of a photoshopped print? Would the equal noise introduced into the whole image be enough to throw off this technique?

  • Joshua Gluck

    It’s an interesting idea. Though in compositing, it’s incredibly important to match grain, especially in film/video. So even for a picture, you could denoise all of the images to a certain extent, then apply a grain pass over the whole thing. I’ve made a habit of filming grain plates every time I have access to a new camera so I can use it later on to add some realistic grain to my work.

    I personally think the other method of checking image compression throughout the entire frame is a slightly more accurate method (especially since many images that are layered in are sourced from various google searches). But it’s definitely getting much harder to tell what’s fake nowadays.

  • David Liang

    Interesting idea but I think technology is already at a point where this can be thwarted easily, I don’t think manipulation can be accurately gauged studying the pixels, I think the future lies in metadata. Adobe, Apple, Capture One, all these guys could add imprints through their software that could give investigators an idea if, and to what extent an image has been altered.

  • Matias Gonua

    it would be easier to use a noise filter in photoshop. Which is what they say on the article, or at least what I understood of it.

  • Samuel

    If i were to see if someone what was photoshopped in my first step would be to zoom right in and see what the noise looked like on both bits, quite often they don’t match even in the slightest. A plugin for this would be useful if you are a forensic or nerd

  • Kevin Lee

    ( •_•)
    ( •_•)>⌐■-■