Finding people with their faces

The story of Daniela Klette went around the world as an example where new facial recognition databanks can unearth identities even for those, who hide for 30 years. A short OSINT review for online investigators.

Techjournalist
6 min readMar 5, 2024
The art of face searching

For the German podcast team, Bellingcat’s Michael Colborne used tools like PimEys and AWS Rekognition to find 30-year militant RAF fugitive Daniella Klette on Facebook images. So much appears straight forward. But, as we learn more about Colborne’s approach, we understand that Colborne with the crew of the Podcast Legion were lucky to search for faces that were long forgotten from the public investigators’ eyes. Also, that others posted and talked about Klette. Klette, under disguise, usually didn’t post much on Facebook about herself. She maintained a fake profile. Mainly image of others unmasked helped to unmask her.

Her posts were about events of the Afro-Brazilian Capoeira association. Sadly, for Klette, the Capoeira association was the weak spot. According to NYT reporter Christopher Schütze, it were those images who got her in trouble. Several lessons we can learn from: First and foremost, in our modern social media society, it may be virtually impossible to stay unphotographed offline all the time. That is, if you want to live a normal life. That also means social engagements. Appearing in the background of others in photos, take at a demonstration or on a train. It becomes exponentially harder to prevent anyone’s face to stay offline. The same is true of our email addresses, usernames, even passwords. In other countries its phone numbers. We leave old pictures on accounts that we long didn’t use or have already forgotten about. Or others leak them online, as we see for Klette.

Soon our voices might sprawl in a similar fashion. All of these are biometric identifiers that AI can scan and associate with our persona. From there, its not far to get access to our whereabouts, social clubs, and homes etc. That said, most of us (certainly Germans) won’t share those details lightly and often it’s not even our fault that they leak. Instead, because there is simply often no way to prevent others — including local authorities, friends, family, businesses, and business partners — to leak such data staying anonymous becomes harder and hunting for targets easier.

Despite its criticism, Pimeyes remains for many of us go-to tool for person-investigations. But rarely use it on its own, but instead often alongside with several other tools and methodologies. Take an example of neo-fascist marching in German streets, offered by image platform Shutterstock.

Were we to find the man in the middle, we could use PimEyes, offering us two more instances where the initial judgement, based on his throat tattoos suggest we could find more images on sites such as eu.zonerama… or le-cdn.website…. If we dont have a full PimEyes membership, we would then use google dork the photo site: “name-of-fugitive” site:eu.zonerama.com. Alternatively, if we don’t know the name, we can use other reverse image tools such as Google Images, TinEye or Google Lense. A wide list, including general reverse image search platforms, can be found here on cqcore excellent Image-Research-OSINT collection. The reverse image search can then help to get more details on a person, as well as finding the original image source.

Tip: don’t discount the power of stock image search platforms. Of these feed platforms such as PimEyes and co, with training images (a number of them listed here)

The German BKA, the Federal Criminal Police Office, has a number of high-quality arrest warrant images on their website. On its own, these don’t serve much hope that PimEyes is able to locate new images, leading investigators into uncharted territory. Mainly because the internet and PimEyes results seem flooded with the very same warrant images. Treatment of such images, such as ageing, can in some instances allow finding better source images.

In Klette’s case, we can use a png face aging app. There are a number of those. One is FaceApp. Some others work better, some give worse results, some are more expensive. The old spiel.

Left the AI aged image of Klette, right the original one

Without PimEyes Deep search, it offers you only a page worth of image results. There is currently no way to find Klettes images of the Afro-Brazilian Capoeira club. This is mainly due to the cramped search results after the hunt for Klette and her partners in crime rejuvenated.

However, we can use a number of facial comparison site, to link her new-old images to her present ones, a crucial step for person verification. There are a number of those, including AWS Recognition, mxface.ai/face-comparing, and others. It offers a confidence score. AWS writes that a 90% confidence suggests a “higher probability of a correct detection or match than 60%”. There is a number of caveats but generally this approach works as a first threshold to ID faces.

PimEyes seems to also disregard hair, and find faces, such as in the case of the Nor­man Vol­ker Franz, who committed a double murder, and broke out of a prison in Hagen, almost 27 years ago. The image by the BKA was taken 1998. Thanks to the police, who frequently help with aging images, we can use one from a police simulation, showing the today 47-year-old.

Left, AI generated image. Right, a simulation by the police.

Though, the results find further images, that are simulations of the fugitive today, it remains a good example of how PimEyes battling age.

An alternative to PimEyes is Social Catfish, allowing reverse image searches. In case of Klette’s aged photo, there are another 56 results. Unfortuntely, Social Catfish charges

To PimEyes there are several, sometimes weaker, sometimes more expensive alternatives. For Russian searches, there is also https://search4faces.com/. We already mentioned Amaozn Rekognition. There is the platform Facecheck, that does not always give conclusive evidence. If you have access to Microsoft Azur, for facial comparison, there is also Micorsoft Facial Recognition.

Klette und co. are probably not the first and certainly the last fugitives that online facial recognition technology finds. Especially interesting for investigators to find targets this way: fugitives still in hiding, for which the police investigation is old enough that traditional search techniques failed, and new technologies have not been applied.

In 2020, there were 481 people from the right spectrum on the fugitive lists with arrest warrants Germany. A handful are from years or even decades ago. Such cold cases could be reopened, now with more advanced tools. We are not at the end of this trail.

--

--

Techjournalist

Investigative journalist with a technical edge, interested in open source investigations, satellite imgs, R, python, AI, data journalism and injustice