Digital photo forensics
|
Jessica Fridrich, a researcher at Binghamton University (and the inventor of speed cubing!) has reportedly filed two patents for technologies that would allow to determine which camera a digital photo has been shot with. Due to tiny irregularities in the production process and in the imaging-chips themselves, every camera produces a random, yet unique noise pattern, the scientists claim. These patterns could be used in forensics to determine whether an image was taken with the same cam, given that two or more pictures or the device itself are available for testing. This would add yet another to the almost traditional links between cameras and guns because it resembles very much the method of ballistic fingerprinting with which the gun a bullet was fired from can be identified. Law enforcement have already expressed their needs for such a software, since it would enable them to identify the sources of pictures on the web to some extent. But just in cases of illegal porn, of course. Related: The cellphone forensics kit. Via Spiegel Online, photo snatched from Flickr and slightly noir-ified. |
2 TrackBacks
Listed below are links to blogs that reference this entry: Digital photo forensics.
TrackBack URL for this entry: http://www.we-make-money-not-art.com/cgi-bin/mt-tb.cgi/6278
[IMGRIGHT=http://www.cellular-news.com/images/press/product_182_a.jpg]Es impresionante lo que se puede saber de algo (y de alguien a su vez) en el rubro "forense" de la tecnología. Por ejemplo tenemos este [URL Read More
I am sure if this is true we will soon see it on CSI or any of the other popular forensic shows on TV today. They claim that every digital camera has a "finger print" which can be used to... Read More
Leave a comment |
|
"every camera produces a random, yet unique noise pattern"
I'm not sure if this has something to do with it, but with software like Neat Image (Nice liitle application that removes noise from a picture), there is the possibility to load up a camera templates that fits your kind of digital cam to make the "cleaning" process more effective. Each camera has a different templates, because each generate a different kind of noise. If this is the same kind of thing, it has been existing for quite some time.
Kiltak
[Geeks Are Sexy] Tech. News
very cool!
This is great as long as they don't modify the photo from its RAW form or post-process it in any way that could smooth or add noise in even the most minor way. What a laugh.
Interesting the analogy between cameras and guns... another great way for "big brother" to control all of us ; )
The inventor is a [deleted for bad manners]. Has she not considered the use of the internet and almost all graphics displayed are in compressed state. The compression distorts and fudges noise levels a great deal. This can only work for RAW images with no compression, you are talking about 1MB per Mega Pixel which no one really uses aside from professional photographers.
Well, I would just say: What if the picture is resized, or recompressed? The pixel patern would be lost.
The US military already does this with other technologies. All electronics can reveal those "tiny irregularities in the production process and in chips". Things like mobile SAM radar stations, cell phones, etc. For instance, Military aircraft patrol the coasts of foreign nations all the time to record locations of those unique signatures and track where those signatures have been, are, and might be in years - to see who's selling who military gear.
Actually, the noise survives lossy compression and typical image processing operations. It would not be that useful if it only worked for raw image formats.
It surprises me how quickly people jump into conclusions and name calling without understanding any technical details about the technology. This is just sad. The work has been peer-reviewed in the top reserach journals. Go study first before putting down somebody's work.
This might offer some hints - but it certainly is no method of broad usage.
I have experiemented with this too: just take a picture (5-10 seconds) with blackened lens (cap works well ;) ) - you will receive a nice picture with digital noise, some pixels are strong (i.e. semi-faulty aka leaking sensor pixels), this really would qualify as a "digital fingerprint" and is reproducable to a certain degree.
Frank, our method uses a different type of noise - the pattern noise. What you described was dark current. In fact, the pattern noise is absent in dark pictures. The dark current was proposed for camera ID by Kurosawa et al. and is of limited use and much less robust to compression or processing.
Well, surely it is more similar with a "from factory water mark", and water marks remain after saving in a lossy format, cropping or similar, even "neat imaging" it, so this finger print could be as sticky as this and survive after publishing the shot in internet.
Jessica, does this pattern exists even in foveon chips like the sigma one or is it related to green-red-blue patterns? Just for curiousity.
Foven has it too. CMOS, JFET, all of them.
The analogy to ballistic fingerprinting is actually pretty interesting - ballistic fingerprints are pretty easy to change, so if someone is worried that their gun is going to link them to a crime, they just need to do a bit of filing. Similarly, if you're worried about illegal images linking to your camera, you just need to muck with the picture a bit, and it would probably be untracable (or, if you prefer, tracable to another camera).
well for those saying that once you compress it the noise goes away. if they were to just take a raw image and compress it then they coudl see the new pattern like this...
image 1 >compress (the original noise is lost but there is a new one)
image 2>compress (same compression should in fact give you same type of noise)