September 12, 2017

Deep Learning: The next generation of digital identification

IDnow is setting new standards for digital identification with the help of deep learning. The Munich-based identity specialist uses image recognition algorithms in their video identification process to help verify identity documents automatically. Using deep learning for this process, IDnow can reliably spot attempted fraud and therefore increase security. IDnow trains its image recognition software using NVIDIA GPUs, which are widely recognised for their capabilities in the sphere of artificial intelligence.

Online verification procedures are currently used by many companies across Europe. These companies require the ability to identify their clients before onboarding them, in accordance with money laundering regulations. During this process, an identity specialist verifies the user’s identification papers via a video call. In addition to human expertise, highly specialized software is used in order to check that both the identity document and the user’s identity are authentic. The software automatically compares images and checks the validity of the ID card using different features like the number of the document.

Digital identities are valuable and readily attract fraudsters. In order to recognise forged and technologically manipulated documents during the identification procedure, IDnow uses deep learning algorithms for image recognition. IDnow developed these algorithms inhouse, as they found that none of the existing technologies, for example those based on the principle of edge detection, gave accurate enough results in practice.

"Video identification takes place in an uncontrolled environment, where the light, background and resolution of the webcam or Smartphone camera used can vary. It can happen that some algorithms detect the edge of an image on the wall instead of the corners of the identity document. Therefore, we developed our own algorithm and trained it in such a way that it can recognise the image of an identity document reliably, regardless of the light, background or camera resolution, and all of this practically in real time," says Armin Bauer, CTO and Managing Director of IDnow.

Next, IDnow’s program classifies the document and detects the country of issue, whether it is an identity card or a passport and the version of the document. The algorithm can also detect and verify data on the identity document, such as the name of the bearer and verification digits. These automatic systems help the identity specialist to detect possible discrepancies or attempted fraud. The image detection algorithms analyse vast amounts of data and slowly learn to recognise connections and patterns and to make appropriate predictions. This training process is being accelerated by NVIDIA GPUs.

IDnow’s technology can also automatically match faces on request. Their algorithm uses 128 points in the face to determine a degree of resemblance between a person in a video call and the person on the identity document. Identification is only possible if the algorithm establishes that it is in fact the same person. This face-matching can also be used to reduce attempted fraud, as known fraudsters can be filtered out at this point in the process.

The IT experts at IDnow have been working on these deep learning models for a while now. As of summer 2016, the algorithms are being used in some clients’ identification procedures. They are currently also being tested for automatic recognition and verification of the security features in identity documents.

"If we can manage to devise algorithms to automatically recognise security features like holograms, which are spread out over a large number of frames in a video, this would be the equivalent to the Holy Grail when it comes to checking identity papers. Our results so far are not just significant in the area of video identification, but also for many other situations in which identity documents need to be checked in uncontrolled environments, for example in the case of driving licences or age checks," says Armin Bauer.

About IDnow GmbH
IDnow provides the world’s most advanced machine-learning technology for its Identity-as-a-Service platform that can verify in real time the identities of more than 5.1 billion people from 65 different countries. Our patent-protected video identification and e-signing solutions help our clients save money, improve customer conversion rates and streamline the onboarding process. We are backed by the two leading venture capital investors BayBG and Seventure Partner and a consortium of well-known business angels. Founded in 2014, we already count a large and fast growing team of 300, making us one of Europe’s fastest growing fintechs. Clients are leading international blue chip companies such as Commerzbank, UBS, Sixt, Erste Bank and Telefonica Deutschland, many fintechs like Fidor, N26 and smava as well as many blockchain-based companies. IDnow was awarded “Most Successful Fintech” in 2017.

Further information available at