How can the authenticity of a photo be confirmed

Image forensics - This is how professionals recognize manipulated photos - and this is how the layman also recognizes them

The three students hold up two posters. One of the pictures on it reads: "School strike for the climate" and "It's our future". In the other picture, on the other hand, the posters say “Electricity and gasoline are not expensive enough - Save the ice berries” and “Get rid of the cars, go on foot”. Otherwise the two pictures are the same. But only one thing is real - photographed by a journalist at the “Friday for Future” demo in Potsdam. The other, on the other hand, was manipulated and distributed by the AfD, among others.

At first glance it is often not easy to recognize a manipulated image. In the case of the striking students, on closer inspection you can see that the lettering on one picture was probably not created by hand, but on the computer.

Google image search can help

The Google image search can also provide a clue. With the "reverse search" function, you can upload an image and see where the image has already been published. For example, you can also check whether a picture was actually created in a completely different context - and, for example, does not show a current, but an event in the past.

But with more and more new technologies, the counterfeits often get better - in such cases, professionals are called for. Jens Kriese is a picture forensic scientist, he checks pictures for authenticity and manipulation. In the interview he explains how it works.

Mr. Kriese, you work as a professional image forensic scientist. As a layperson, do you have no chance of recognizing manipulated photos these days?

In many cases not. A study has shown that only a fraction of the manipulated images on Twitter, Facebook and Co are recognized by image forensics. This is due, among other things, to the reduction to the web format. When compressing large images for the web, certain “brush strokes” can be masked.

But why are we so bad at detecting tampering?

This is a product of our long evolution, which was aimed at trusting the visual impression. When our ancestors saw an attacker, they had to flee and not think: “It won't be that bad.” Pictures still enjoy this trust today.

How do professionals proceed in such cases?

Once an image has been edited with a program, it contains a kind of invisible fingerprint. So if someone wants to sell a picture as original, but their fingerprint points to an image processing program such as Adobe instead of a smartphone or a reflex camera, one should be skeptical.

Jens Kriese works as a picture forensic scientist in Hamburg. Source: private

Which technologies are then used?

These are mathematical methods, such as statistical measures. Take the following example: If you photograph the same scene twice with your smartphone - once in sunshine, once at night, you will notice that the night shot is more jittery, it has more noise. A different noise behavior can not only be seen in the comparison of day and night recordings, but this is also specific, for example, for different cameras or the sensitivity setting. So if someone creates a collage from several images from different sources, programs can ideally discover the different noise. It becomes more difficult if the collage was created from images from the same source.

With the help of artificial intelligence, images can now not only be manipulated, but also created from scratch. The results are deceptively real to humans. Isn't that a big problem?

We have AI on both sides of the scale, not just on the side of tricksters and propagandists. The neural networks can therefore also be used to increase the selectivity and, in turn, to better identify forgeries. In addition, an AI may also leave digital traces. It is already similar today, for example in the cinema, when scenes are not filmed but are generated by a computer. These images then have different characteristics than those of a camera, for example.

Read here:
Deepfakes: Are Manipulated Videos a Danger?

But that also sounds like permanent competition with the counterfeiters. Do you have to learn new tricks all the time?

Yes, the host of methods is growing. There isn't one approach that always works. Because these processes cannot be automated, an image forensic scientist must therefore know exactly with which question he is approaching the image. But you also have to say: Image forensics can never confirm whether a picture is really real - it could also be a perfect fake. It only becomes really helpful when the forger has made mistakes, when something is inconsistent. This usually happens under time pressure, when a fake is knitted with a hot needle and then, for example, the lighting conditions are inconsistent or the perspective is wrong.

By RND / Anna Schughart