Multiverse 2.357

Photographs have clues that are specific to a camera. Tiny clues, but given a group of photos form the same camera it quickly becomes obvious that they did come from the same source. Once that happens, every photo you ever take is identified as coming from your camera, on your phone, and associated to you.

Most readers at this point would probably think, “Why should I care?” The answer is that this does not just affect investigative journalists. It will affect us all because as AI systems get more capable, they can infer an amazing amount about your life from a collection of seemingly unrelated facts. Bots scour the Internet now, gathering and organizing collections of photos. The current AI technology is not advanced enough to definitively associate all photos to the people who took them, for various reasons. These reasons are just temporary technical roadblocks – they will be overcome in time.

One very sophisticated method of demonstrating that camera images came from the same device has to do with patterns of noise generated by heat in that device. Noise signatures in photographs come from ambient thermal heat, and thermal heat generated in the camera – inside the charged coupled device (CCD) itself.

Astrophysicists have long been interested in this, but the easiest solution for them is supercooling the camera, which is not an option for a mobile phone user. Camera manufacturers have built products that heat the CCD just before taking a picture, but to date this is not available in mobile phones.

Mobile phone cameras product images that display patterns of random noise in the least significant bits (LSB), which basically means the colors are altered only a tiny amount. This alteration is small, so that a human eye can’t detect it but computers can see these patterns and easily determine that it came from a certain camera.

Unfortunately, for those without a forensics lab, the easiest way to make this difficult to detect is to degrade the image. A quick and easy way to make this signature less obvious would be to save in a lossy format like JPEG, then crop or resize and save again. The obvious problem is that this is noticeably making your photos look worse.

Moreover, you would need to vary this process of saving, altering and saving each time to avoid simply changing the pattern to be detected. As if that weren’t daunting enough already, you probably need to save with fairly high compression so that it really is a lossy process. That is not something you can completely mitigate without extreme effort, but in the
process of avoiding simpler techniques you can make this more difficult.

So with that out of the way, let’s look at simpler ways of determining that photos come from the same camera; ways that you have more control over. Dust and scratches are an easy ways to determine that a group of images all came from the same camera. Tiny specs of dust unnoticeable to the eye can prevent certain pixels from changing in any of your photographs. Facebook has patents for identifying a camera based on faulty pixel positions caused by dust, scratches and similar conditions.

Clean the lens of the phone’s cameras with a cleaning cloth made for cameras or eyeglasses to increase your chances of avoiding this issue. Also try to take pictures of a wider physical range than you actually need, and crop the image in the processing phase. Other techniques that can help are resizing the image and adding adjustment layers – both can alter the pixels to avoid those dust dead spots.

This post is a followup to the previous post Cameras, People and Photos. I have a lot more to say about this topic, so stay tuned for the next installment where I want to take a closer look at JPEG files, EXIF data and digital photographs.