PRNU Based Verification of Multi-Camera Smartphones

Manoranjan Mohanty: Hi, thanks for watching this video. This is about our short paper titled ‘Verification of Multi-Camera Smartphones, which was done in collaboration with UTS and University of Auckland.

So basically PRNU-based source camera attribution is a method which can tell if a particular anonymous image has been taken by a particular camera. This is really useful for crime images, which have been shown here as an example, and these images can come from social media.

So, as we know, the number of crime images in social media are increasing and in that case, it could be really nice to know if a particular anonymous image, which is crime in nature has been taken by a particular camera or not.

And the PRNU-based method is one of these methods which can help us in this. The way PRNU works is very simple. So, each camera has a sensor, which we know as a camera sensor, and the way this sensor has been built in such a way that the sensor always has a fingerprint. This is particularly a type of noise which is unique to a particular camera sensor, irrespective of the camera model.

The way we find that fingerprint is also very simple. We just need to know a few images of this particular camera sensor, let’s say 20 or 30 images, and from these images we compute what we call as “noise” from each of these images and then combine the noise, then we get a fingerprint or we get a combined noise, which we call as the PRNU noise, which is unique to a particular camera sensor irrespective of document or model.

And therefore it becomes a fingerprint that we get and then store in a database. And in the latter stage, when we have an anonymous image, and the question we’re going to answer is whether this particular image has been taken by this particular camera or not.

Then similarly, we get a noise and then match the noise with document frame that we had compared before. And if that is a match, then we say that this anonymous image has been taken by this particular camera.

Now, PRNU-based method is not a new method. It has been studied for years, mainly for single camera. But as we know, the number of multi-cameras are increasing nowadays. All the tough cameras, mainly smartphone cameras, are now multi-cameras. And these multi-cameras are being used for multiple properties, mainly to increase the quality of the image that we are getting from the camera, mainly smartphone camera, either for optical image, optical zooming, [inaudible] pic, and so on.

So the objective in this particular work was to find out whether PRNU-based method works for this particular camera, which is basically multi-camera. And if so, then what is the best way to find the fingerprint and do the matching? And we found that there is no reason that PRNU-based method will not work for multiple camera smartphones and also, actually explore two different methods to do the PRNU method on multiple camera smartphones, these are listed here.

The first one is known as ‘multi-fingerprint approach’ where basically we compute a fingerprint from each of these cameras in the multi-camera smartphones. So these have two different sensors when we’re talking about two different cameras in the multi-camera thing, and then we have to get a fingerprint and then we store them in the database. And when there is an anonymous image, we get the noise, and then we match this noise with each of these fingerprints to find out if, you know, this noise is matching at least one of them.

And if so, then we conclude that this anonymous image had been taken by this camera. The other one here, what we do is we get the fingerprint from each of these cameras, but instead of storing two different fingerprints, we combine them. And then we store the combined fingerprints which we call a “mixed-fingerprint” and during the time of matching, then we get the noise and then over the correlation, find out if the noise is matching to the mixed-fingerprint.

So these are two different approaches that we explored in this particular work. We do an experiment for multiple multi-camera smartphones, and this is still in the preliminary stage, but from the data that we have got so far, it looks like for some cameras, multi-fingerprint approach works better, on some others mixed-fingerprint approach works better. But there are also some cases where the PRNU method is not working on a few of these cameras and we are still exploring what could be the reason. Yeah. So that’s what this paper is all about. Thank you.

Leave a Comment