Information Security Stack Exchange is a question and answer site for information security professionals. It only takes a minute to sign up.
Sign up to join this community
It is well documented that one should consider that all modern printers perform steganography / watermarking in the back of their users.
Is such a feature also documented / are there rumors of similar features with the raw pixel matrix coming out of modern digital cameras?
I am not talking about exif of other metadata, only raw pixel data. I am also not talking about naive steganography as done by adding dots, but also advanced methods, as discussed in the comment section.
Yes - well, sort of, albeit (as far as I know) unwittingly. I attended a conference where one of the presenters was prof. Jessica Fridrich reporting on the identification of digital image sources. You can probably start your search on her page, here.
To be clear, this is nothing like intentional yellow-dotting performed by some printers, and cannot be used to trace a picture to a camera unless you also possess that camera and can take other pictures to compare.
What happens is that every CCD chip has subtle variations that can be recovered through frequency analysis from the original picture data, and these variations are manufacturer and chip specific. This allows, with a certain error margin, to:
given two pictures, determine whether certain areas of one picture are original or whether they come from a different source (the patterning in those areas is not consistent with the CCD)
given two pictures, or a picture and a camera, determine whether the picture came from that camera or not.
It would be extremely easy to make this system more reliable, by e.g. laser etching the CCD chips with a unique pattern. This would not be visible to the naked eye when inspecting the pictures, but the information could be recovered.
Phone picture watermarking
Cell phones are able to take a picture as well, using camera modules that usually output a RGB or HDR matrix. This is then encoded in a suitable file format (usually JPEG) by software on the phone.
For all the purposes I can think of, what is needed for watermarking is to be able to know that a given picture was actually taken by a specific phone, usually because that is proof or very strong indication that the photo was taken by a specific person, or in a certain area, or such. The image having been taken by a specific CCD (maybe originally mounted in a different phone) is irrelevant. Actually, even if I could prove beyond any doubt that a picture was taken using Joe Q. Average's old CCD camera, it would avail me nothing if it could also be proved that, when this happened, Joe's CCD was actually inside my phone. The clinching proof here involves the phone.
At the same time, CCD watermarking only requires (and needs) to have the phone available. Whatever CCD is in the phone is irrelevant - and if I could have replaced the CCD, I might as well have replaced the phone. Being able to identify the CCD provides no advantage.
So, picture watermarking makes perfect sense as long as it is done by the phone, on whatever data comes from the camera. Embedding such a watermark in the picture can be done in several ways (I worked with some of them), depending on how robust the WM needs to be. For example: I can Fourier-transform the picture, getting myself a bunch of frequency amplitudes. I can then quantize (some of) those frequencies, or the correlation thereof, using QIM or other methods, and distribute the watermark among those quantizations in the form of bit pools. Finally, I reverse the transform, obtaining a new image that is visually indistinguishable from the original and, lacking the decoding key, has no obvious telltales of having been watermarked (not always true - the 'remainder' distribution of the frequencies might show a 'clipped' or 'clustered' aspect, which, while not telling me what the watermark is, would be a strong indication that a watermark is there - or at least, some weirder-than-usual signal processing took place). This watermark can still be recovered if the image is clipped or cropped, rotated, and shrunk/enlarged. With more difficulty, it can be recovered after the image has been slightly blurred.
Since this technology exists, embedding a watermark in the CCD hardware rather than in the phone firmware seems both useless and impractical; it would maybe have the advantage of not being a software artifact that might be detected and reverse engineered, but modern phones already have tamper-proof kernels (for example, the iPhone uses the Secure Enclave processor and memory for sensitive operations. It would not be difficult to have the SE chip watermark each and every picture, and the code doing this would be hidden inside the Secure Enclave).
