Thursday, 4 June 2015

"how to not do image analysis." What Bellingcat is doing is nothing more than reading tea leaves.

SPIEGEL ONLINE: Bellingcat made headlines around the world this week when it claimed on Sunday night it had proven that Russia's Defense Ministry conducted forensic manipulations. The allegation is focused on images of the shooting down of Malaysia Airlines Flt. MH17 in eastern Ukraine last spring.

Kriese: The term "forensic analysis" is not a protected one. From the perspective of forensics, the Bellingcat approach is not very robust. The core of what they are doing is based on so-called Error Level Analysis (ELA). The method is subjective and not based entirely on science. This is why there is not a single scientific paper that addresses it.

SPIEGEL ONLINE: What's the hitch?

Kriese: Forensic scientists use computer procedures that allow for the clearest possible conclusions: Has it been manipulated -- yes or no? Contrary to what Bellingcat claims, Error Level Analysis does not provide clear results. The conclusion is always based on the perspective of humans, on their interpretation.
SPIEGEL ONLINE: What does the method entail?
Kriese: It attempts to determine compression artifacts. Those are the small deviations created when a photo is saved in JPG format -- differences from the original. It is possible to depict them in color. But: The final decision on whether a manipulation has occured or not is then still a personal decision made by the viewer. One has to decide whether variations should be attributed to manipulations or are they normal and could be attributed to clouds, for example?
SPIEGEL ONLINE: Do you consider the Russian satellite images to have been manipulation?
Kriese: That's not the right question. We are not talking about satellite images here. We only know the version published by Moscow. That is a satellite image that has been prepared for use in a presentation.
SPIEGEL ONLINE: Bellingcat has come to the conclusion that they were edited using Photoshop.
Kriese: That's an erroneous interpretation. They claim that the metadata shows that the images were processed using Photoshop. Based on that they are concluding it was the clouds that were likely added in order to conceal something. The truth is that the indication of Photoshop in the metadata doesn't prove anything. Of course the Russians had to use some sort of program in order to process the satellite image for the presentation. They added frames and text blocks in order to explain it to the public. The artifacts which have been identified could be a product of that -- or also a product of saving multiple times in JPG format.
SPIEGEL ONLINE: Bellingcat says its findings are based on the use of the analysis tool, a website.
Kriese: And its founder Neal Krawetz also distanced himself from Bellingcat's conclusions on Twitter. He described it as a good example of "how to not do image analysis." What Bellingcat is doing is nothing more than reading tea leaves. Error Level Analysis is a method used by hobbyists.
SPIEGEL ONLINE: How could one really test whether the satellite images have in fact actually been manipulated?
Kriese: That is very difficult. Ideally it would require the original documents, the satellite images themselves or perhaps even the raw data. The Russians have them.
SPIEGEL ONLINE: Can't raw data also be manipulated?
Kriese: That is laborious. Other methods are more effective. There's an entire discipline exploring how image manipulation can be concealed. It is called anti-forensics. It allows photos to be sharpened after they are taken or to be edited with blur filters. Ninety percent of the time, ambitious bloggers like Bellingcat get nowhere.
SPIEGEL ONLINE: So how is it even possible to uncover falsifications?
Kriese: Most are created under time pressure, which leads to small mistakes. But no one would be reckless enough to use Photoshop of all things and then not clean up the metadata. There are very different variants, and I think the intelligence services know a few of them.
SPIEGEL ONLINE: Satellite images are often used as proof of events in the Ukraine crisis, even by NATO. Are they even meaningful?
Kriese: It is easy to claim to amateurs that one can see this or that. But just think about the US images of the alleged poison gas facilities in the Middle East. There's a similar point at Bellingcat: In one of the photos, a growing spot can be seen. It's allegedly an oil puddle next to a vehicle. But does one consider that to be plausible? I think it depends on whether a person wants to believe it or not.
SPIEGEL ONLINE: What's your personal conclusion about the satellite images?
Kriese: The first thing to die in war is the truth. Each side likes to throw random smoke bombs. There is no way of knowing if the images show what Moscow is claiming. What one can say, however, is that this "analysis" has achieved nothing besides raising awareness of Bellingcat.
Zur Person

  • Jens Kriese
    Jens Kriese studied biology at Justus Liebig University in Giessen, Germany, and worked as a scientific researcher and in the field of e-learning. He developed the stock photo archive IRISPIX and is a professional image analyst. He owns an office for digital imaging forensics in Hamburg.

  1.   retweeted
    Take some of your own Medicine : Twitter Users Find Ukrainian Buk-M1 System After Kiev Denies Existence

No comments:

Post a Comment

Note: only a member of this blog may post a comment.