Pixel noise said to create unique camera "fingerprint"

Researchers at Binghamton University in New York have filed for a patent on a method of extracting unique identifying information from digital photos, based on pixel noise. According to the researchers, each camera generates its own digital "fingerprint" based on random variables that occur in the manufacturing process. By comparing multiple images from a single camera, the researchers say they can trace the images back to the specific camera that took the photos. The developers see potential for the method in law enforcement activities, such as tracking down child pornographers and forgers. We hope they're right; certainly, any tools they can use are worth exploring. However, we can't help but wonder whether clever criminals will just be able to run a few Photoshop filters and be able to wipe out any trace of their cam's fingerprint before uploading their pics.

[Via Ars Technica]

[Korey @ Apr 23rd 2006 10:44AM]

I live about 1/4 from BU Mad Props to em now lets get hard evidence on Michael Jackson

[Kim Laughton @ Apr 23rd 2006 10:49AM]

I expect any resized photos would lose this 'fingerprint', so that applies to virtually anything online.

Might be more useful for noise reduction?

[Spoonman @ Apr 23rd 2006 11:02AM]

Someone's already using it for noise reduction:

http://aggregate.org/DIT/peepfish/

[Chris McDowell @ Apr 23rd 2006 11:30AM]

It has been hacked before it was even implemented. http://hackaday.com/entry/1234000793073646/ or just add in or reduce your own noise.

[Mark @ Apr 23rd 2006 11:53AM]

"Clever criminals" is generally an oxymoron. Criminals tend to be idiots; that's what leads them to a life of crime in the first place.

If the technique does prove viable, I'd expect another use of it by law enforcement agencies: to track down the anonymous someones who snapped shots of police brutality and the like. Even the thought that the government could find you based on your digital camera's "signature" could have a chilling effect.

[IraqiGeek @ Apr 23rd 2006 11:56AM]

jpeg has become the standard image compression method used on the net, and by definition, its a lossy compression method. Forget about resizing. Just compress the image to "normal" jpeg settings, and that noise info will be lost rendering the image virtually untracable. In fact, that info is already lost when you download those images from your camera, as it would have already compressed the images before saving them on the flash memory loosing all that noise information.

This method may work on raw/uncompressed images, but how many people will post huge, multi megabyte raw images online???

[Jason Huggett @ Apr 23rd 2006 12:25PM]

Looking at it from the other angle, it could provide useful for evidence photos. A method in which you can prove that the photo evidence hasn't been tampered with. I would imagine you would have to use a lossless format though.

[Mike K. @ Apr 23rd 2006 12:29PM]

This isn't news to astronomers. Ever since the arrival of CCD cameras, astronomers have had to deal with CCD "fingerprints" such as the type this article is talking about. You don't notice them when you're taking images in the daylight, since the photon count per pixel is so high.

However, once you start talking about deep-sky images, suddenly the underlying noise structure on the chip becomes a problem. Since no two CCDs are alike, you have to calibrate your images for each camera you use. A quick google for bias & fark frames, and flat fields will turn up all kinds of info on the techniques used by astronomers to deal with this stuff.

[Josh Warner @ Apr 23rd 2006 12:40PM]

I have a Nikon 5700, and it has a built-in feature called, cryptically, "Noise Reduction" :)

When enabled, it takes one shot with the shutter open normally, and then can take one or two more with the shutter CLOSED, simply generating the characteristic noise pattern. The camera then subtracts the noise from the photo automatically, and saves the resulting
"cleaned" image. This is especially helpful for long exposure night shots, when noise becomes a real issue, and it works wonders.

I don't think this is a widespread feature in consumer cams (yet), but I think that this kind of feature might render this method of identification next to useless. At the least, it should be more difficult to extract a noise "signal" from these shots.

[Mike K. @ Apr 23rd 2006 12:40PM]

Heh. That would be bias & DARK frames. :)

Whoops!

[Moogle @ Apr 23rd 2006 1:26PM]

Yeah, noise reduction can hinder this kind of tool, but it's not generally used because it takes longer and is only helpful in a few situations and with low/no compression. It's still a useful tool to have in ones toolbox.

Compression will wipe out the exact CCD signature, but it doesn't eliminate any signature altogether. It might require a larger sample size (number of shots) from the camera to get a good margin of error, but it's probable that information can still be extracted that's similar from jpeg to jpeg that the camera creates. This is of course, avoidable too. You could have a program randomly scale every image and recompress them a few times at a few resolutions before cutting them down to a final size.

Also, most criminals commit crime because they're too dumb to get what they want in a legal way. Most pedophiles can't get what they want legally, so there are bound to be smart ones out there as well. Plenty of stupid ones, sure. But you'll never be able to catch all of them. That is, unless we go into total intrusive society mode, and I'm more opposed to that then having very few minimally intrusive pedos in society (they have to make as little noise as possible to stay hidden). It sucks, but it's not worth the price, IMHO.

[Chris Bell @ Apr 23rd 2006 2:10PM]

This would be more useful for proving cases in court because unless manufacturers start maintaining a noise profile database of all cameras they sell, then law enforcement will need to have the camera on hand to tell where a noise profile originated.

As for the noise itself, it's not enough to simply reduce the size of the photo or compress it. The noise pattern still exists. Noise reduction might work but it's more difficult to do with the quality you are likely to have with consumer level cameras. Chroma noise (looks like red and blue splotches) in particular is hard to remove even with the best tools. It can be elimimated almost like magic on RAW files taken with pro level cameras.

Perhaps adding noise would do the trick?

[Chris @ Apr 23rd 2006 3:16PM]

Couldn't a criminal simply change cameras periodically so older photos would not be traced back to him?

[Karl Viklund @ Apr 23rd 2006 3:32PM]

"However, we can't help but wonder whether clever criminals will just be able to run a few Photoshop filters and be able to wipe out any trace of their cam's fingerprint before uploading their pics."

Well. Maybe. But I still see a future for this technology. There are alot of bad stuff out there already which they can trace with this. And all criminals are not and will not be smart enough. This is just one of many technologies that can be used to fight child porn.

[Foof @ Apr 23rd 2006 4:36PM]

Josh,

Some form of noise reduction is used in almost all consumer digicams, including dark frame subtraction for cameras that support shutter speeds over 1 second. Now, I'd imagine this kind of processing could "see through" standard jpeg or dark frame processing, particularly as dark frame results in just as unique a signature, even if your eyes do not perceive it as "noise."

However, the article is correct in that a single filter, such as a noise->median at lowest radius, would obliterate any identifying noise characteristics. Even more, as some perviously mentioned, simply resizing the image for web would do the job with 100% confidence.

Only criminals who upload original-sized 5-10MP pictures need worry, and if they're already doing that, there's probably more than enough information in the EXIF alone (which they would not be aware of) that would tie any shot back to their camera (Name, time, shooting settings, shutter actuations, etc.).

[CeeJay @ Apr 23rd 2006 6:14PM]

If you can extract the noise-fingerprint from a camera , then you would also be able to fake someone elses camera by using by subtracting the noiseprint of your own camera and adding the noiseprint of the camera you wish to pass as using.

"Your honor - I've been setup !"

[Ryan Gardner @ Apr 23rd 2006 6:57PM]

GrainSurgery - a photoshop plugin - can do wonders with grain. It can suck out just about all of the grain, depending on how much you let it do... and it can also add grain to make it look like it was shot on different kinds of film.

I would be very suprised if removing the grain and then adding in a (subtle) grain pattern similar to Kodak Kodachrome would leave anything of the fingerprint behind.

[Magallanes @ Apr 23rd 2006 10:58PM]

There was not need to hide the "fingerprints" because it's hard to identify who's the owner of the camera.. in a very optimist viewpoint you can determine the mark and model of the camera but currently many cameras do it directly.

But what's happen for a dirt lens?. For example someone touch the lens. It can create some pattern in the photo but after, the photographer can clean the lens so the pattern will change. In other word :a true fingerprint (if the dirt is because fingers) can generate digital fingerprint?

[Drew Noakes @ Apr 24th 2006 6:20AM]

This made me recall...

My brother sells cameras for a living, and told me that the police association buys cameras that burn directly onto CD (rather than memory sticks) as it's harder to tamper with the digital evidence.

Not exactly in line with the article, but an interesting and related point.

[TomK @ Apr 24th 2006 6:21PM]

As much as I hate "relevance-through-proximity" commenting as we've sometimes seen here (re: http://www.engadget.com/2005/03/08/how-to-make-your-own-annotated-multimedia-google-map/ ) .. I grew up in the Southern Tier of NY and have always felt SUNY Bingo was underrated as a state school, especially regarding national semblance and tech research as in this article. However, any article that brings them noise is cool in my book ... as Korey so eloquently put it ... "Props!" :)

[Tom @ May 4th 2006 8:41PM]

#18 (Magallanes), They're not talking about finger prints on the lense, they're talking about a "digital 'fingerprint' based on random variables that occur in the manufacturing process." Please read things.

But anyways, by making reference to certain top-secret Photoshop filters, engadget is essentially giving child pornographers the information they need to clean up their tracks! Can't they see they're giving aide and comfort to our enemies?!