Camera model identification based on forensic traces(sciencedirect.com)
sciencedirect.com
Camera model identification based on forensic traces
https://www.sciencedirect.com/science/article/pii/S0957417422010430
16 コメント
I think an interesting possible application of this is that a lot of applications of video to 3D reconstruction, such as SLAM and photogrammetry, require good camera intrinsics as calibration input. Getting this information automatically seems to be difficult, so lens calibration is an important step. This work suggests to me that calibration information could either be recovered directly from images, or the correct entry could be automatically looked up in some large shared database, which would really simplify things for certain applications.
Can an image be completely re-created with an AI routine or process that yields very high visual fidelity to the eye but eliminates any camera signatures?
Wow. Good find! Almost exactly what I was imagining.
I'd say also if this interests you - checkout #OSINTION #blackbadge training from Joe Gray! https://www.theosintion.com/courses/ (also, I am not Joe Gray - just lots to be learned from this category of investigation)
I could have sworn I read about this exact thing in Cory Doctorow's novel Little Brother
How does that help? It’s like identifying in which continent a picture was taken.
Hardly. There's 7 continents (~3 bits of entropy), but dozens/hundreds of camera models this technique (6-8 bits). And it's hardly the only media forensic technique out there. There's ways of extracting the time of day without metadata, manipulation software chain, the lens used (sometimes down to make and model), and tons more. Before you know it, you've put a big dent in the ~33 bits needed to pinpoint a specific person.
I'm sure it will be useful for law enforcement or a defense attorney, if someone is on the record saying "I took this image with my Model Z camera" and they can prove that's a lie, it brings the rest of that persons testimony into doubt.
I expect it's similar to knowing the make and model of a car that was used in a crime.
Police aren't going to go around pulling over every one of those cars. However, if a suspect is captured, it can be a datapoint to help make the case.
Police aren't going to go around pulling over every one of those cars. However, if a suspect is captured, it can be a datapoint to help make the case.
> However, if a suspect is captured, it can be a datapoint to help make the case.
The paper talks about using this to go after people who are creating CSAM. That's a federal crime, which means an investigation will be conducted by federal agents. USAs have a 93% conviction rate – they'll case build like crazy to ensure they get a conviction. Any data point helps.
Plus, they might ask you if you took a photo with your phone – say "no" and they prove you did, and you're also lying to the feds, which is a felony itself.
The paper talks about using this to go after people who are creating CSAM. That's a federal crime, which means an investigation will be conducted by federal agents. USAs have a 93% conviction rate – they'll case build like crazy to ensure they get a conviction. Any data point helps.
Plus, they might ask you if you took a photo with your phone – say "no" and they prove you did, and you're also lying to the feds, which is a felony itself.
There are no single perfect solutions. Being able to identify which continent a photo was taken in would often be useful. Geolocation, even inexact geolocation, is an essential part of intelligence work.
my point is that there are hundred of millions of users using the same iphone camera. That looks like a very minor datapoint to me.
It's a point of correlation. Suppose you know person X in your area has a tattoo of a skull on their right arm, smokes Marlboro red cigarettes, is left-handed, and uses an iPhone, and all this information is fairly recent. You notice someone with a skull tattoo and decide to observe them, the three minor details take the id from a possible to a probable.
True - but I might be the only person in my county using a Fuji X-Pro3.