The different sensitivities of the photosites creates a type of imperceptible image watermark. Although unintentional, it acts like a fingerprint, unique to your camera’s sensor, which is imprinted onto every photo you take. Much like snowflakes, no two imaging sensors are alike.
In the digital image forensics community, this sensor fingerprint is known as "photo response non-uniformity". And it's "difficult to remove even when one tries", says Jessica Fridrich of Binghamton University in New York state. It's inherent to the sensor, as opposed to measures, such as photo metadata, that are "intentionally implemented", she explains.
The upside of the non-uniformity fingerprint technique is that it can help researchers like Fridrich identify faked images.
In principle, photos constitute a rich reference to the physical world, and can therefore be used for their evidential value, since they portray what is. However, in the current climate of disinformation – exacerbated by the ready supply of image editing software – it has become increasingly important to know the origin, integrity and nature of digital images.
Digital camera photosensors contain tiny imperfections that act as a fingerprint (Credit: Getty Images)
Fridrich has patented the photo fingerprinting technique, and it has been officially approved for use as forensic evidence in court cases in the United States. It means investigators can identify manipulated areas, associate it with a specific camera device, or establish its processing history.
Fridrich believes this technology could also be used to reveal AI-generated synthetic imagery known as deepfakes. And tentative research corroborates this. The distinguishing feature of a deepfake is its photorealism. Having gained infamy in 2018, due to their use in pornographic videos, deepfakes present a tangible threat to the information ecosystem. If we are unable to differentiate between what is real and what is not, then all media consumed can be reasonably doubted.
In the post-truth age, the ability to spot fakery is obviously a positive development. But at the same time such photo-fingerprinting methods can "have positive and negative uses", says Hany Farid, a professor in electrical engineering and computer sciences at the University of California, Berkeley and founder of digital image forensics.
While Farid has used the non-uniformity technique to link photos back to specific cameras in child sexual abuse cases – a clear benefit – he also cautions that as "with any identification technology, care should be taken to make sure that it is not misused". This is particularly pertinent to individuals such as human rights' activists, photojournalists and whistle-blowers, whose safety may depend upon their anonymity. According to Farid, such individuals could be "targeted by linking an image back to their device or previously posted [online] images".
Reducing image resolution could make the fingerprint less detectable (Credit: Javier Hirschfeld)
When considering these privacy issues, we might draw parallels with another technology. Many colour printers add secret tracking dots to documents: virtually invisible yellow dots that reveal a printer's serial number, as well as the date and time a document was printed. In 2017, these dots may have been used by the FBI in the identification of Reality Winner as the source of a leaked National Security Agency document, which detailed alleged Russian interference in the 2016 US presidential election.
Regardless of your opinion on whistleblowing, these surveillance techniques could affect us all. The European Commission has voiced concerns, suggesting that such mechanisms could erode an individual's "right to privacy and private life". If we view photo fingerprints as being equivalent to a printer's serial number, then this prompts us to ask whether photo response non-uniformity also violates an individual's right to protection of their personal data.
Despite our chronic predisposition to self-disclose over the internet, we vehemently reserve the right to privacy. In principle, people should be able to decide the degree to which information about themselves is communicated externally. But in light of what we now know about forensic photo tracking, such self-determination may only be an illusion of control.
Standard metadata is difficult enough to avoid – you have to scrub it afterwards, and the only piece of information you can stop from being created in the first instance is photo geolocation. Photo response non-uniformity, however, is far more difficult to extricate. Technically, it should be possible to suppress, for example, by reducing the image resolution, says Farid. But, by how much? This of course depends on many factors such as the type of device used for image capture, as well as the fingerprint matching algorithm employed. There is no one-size-fits-all solution to fingerprint removal.
So, how concerned should we be about photo response non-uniformity from an ethical standpoint? When I asked Fridrich about the implications of its various applications, she candidly remarked, "a carpenter can do wonders with a hammer, but a hammer can also kill". While no one is saying that the hidden data inside your photos could be deadly, her point is that this is a technique that could cause harm in the wrong hands.
You don't need to be Donald Trump or John McAfee to be affected by the rise of photo metadata and fingerprints. So the next time you take a snap with your smartphone, you might pause to reflect on how much more is being captured than what you see through the lens.
*Jerone Andrews wrote this article while working as a researcher at University College London, as part of a media fellowship organised by the British Science Association.