Securing the Image

Nasir Memon, Principal Investigator of NYU Abu Dhabi's Center for Interdisciplinary Studies in Security and Privacy researches methods photographers can use to anonymize their images.

From selfies to war photography, images are increasingly born digital. But as photographers snap and share, most don't realize that the images they post online can be traced back to them using digital forensics. If you're sharing a photo of your brunch on Instagram, this is of little concern. But if you're photographing in a war zone, the stakes are higher.

Citizen-journalists are a main source of information that comes out of conflicts — the civil war in Syria is testament to this — and digital photography and video are key modes of their communication. In some cases, these reporters go to great lengths to maintain their anonymity. But their tools can reveal their identity, as every digital camera has its own unique noise pattern, or "digital fingerprint."

There are two sides to the issue. Digital forensics provide law enforcement an important instrument in their fight against crime. But these techniques can also leave activists and reporters vulnerable. "People expose atrocities by sending pictures," Professor Nasir Memon said. "And if those pictures can be connected to a particular camera and thus to a particular individual, that would not be a good thing." In places like Syria, where citizen-journalists have worked to expose the brutality of war, the only way to ensure anonymity is to remove this noise pattern.

Memon is professor of computer science and engineering at NYU School of Engineering and principal investigator of NYU Abu Dhabi's Center for Interdisciplinary Studies in Security and Privacy (CRISSP-AD). He and his colleagues at CRISSP-AD recently published a paper that describes techniques that may provide tools to help photographers anonymize their images.

Instead of capturing light on film, digital cameras capture photons on a photo-reactive sensor that is made up of millions of smaller sensors. For example, if you have an eight-megapixel camera, the image sensor is made up of roughly eight million smaller photo-reactive sensors (a pixel in a digital photo corresponds to an individual photo-reactive sensor). Since manufacturing processes aren't perfect, the photo-reactivity of each tiny sensor varies; one may be slightly more or less photo-reactive than the one next to it. This minute variability from pixel to pixel gives each individual camera its "digital fingerprint." In technical language, the pixel pattern is called the photo-response non-uniformity (PRNU) noise pattern.

So, if a digital forensics specialist knows that a photo was taken with a particular camera, the specialist could compare the pixel pattern of that image to other images and determine if those images were taken with the same camera, simply by looking at the pixel pattern. It is a complex process, but with the right tools, it's possible.

Memon and his colleagues "have applied for a patent for a technique that has proven to be very effective for removing the noise pattern of a particular photo. And this is not only in terms of practice, which can be shown by experimentation, but even by mathematical analysis," so that the digital fingerprint has been completely removed from the photo.

In time this technology could be used to filter an image after processing, or it could be integrated into a social media platform and be applied to a photo before the image is posted on the Web. A tool like this could help protect the anonymity of the photographers who need it most — brunching shutterbugs included.