Identification Document Liveness Detection

From Antispoofing Wiki

ID liveness detection has become vital for the digital onboarding process as more companies, services and institutions need to verify user identities remotely.

ID Liveness Detection: General Overview

Document liveness detection refers to proofing that a document presented remotely is authentic and real. This is necessary to verify people, detect forgeries, as well as prevent potential digital crime. There are various types of fraud that involve fake document usage online.

They include identity theft, money laundering, illegal purchasing of restricted goods or prescribed medicine, accessing private data, and so on. Terrorism assistance and sponsoring is one of the critical threats too, as by using a physical fake ID, an attacker can purchase plane or public event tickets online and stay unnoticed for the authorities.

ID-based attacks also vary in ways and means. A type called Presentation Attack (PA) implies that a malicious actor can show a printed or digital photo of someone else’s likeness to the sensors if the remote identity proofing (RIDP) requires a selfie to complete authorization.

Face morphing, while not being an attack type on its own, is a method gaining popularity among malicious actors. By combining facial features of two or more people, they can produce an ID photo or fake selfie that has a chance to be authorized as both/multiple applicants.



Another attack directly involves fake, stolen or expired documents. To detect them, a document liveness check is requisite. It analyzes various properties of an ID presented, including its 3D properties, expiration/issuance date, coat of arms, holograms, stamps and other security details.

A noteworthy case of online ID fraud took place in 2020 when a perpetrator used multiple accounts, fake driver licences and wigs to spoof ID.me — the digital identity verification platform employed by the US government agencies. According to their report, $900,000 were claimed on the unemployment grounds before the scam was detected. The eventual goal of the fraud was to steal $2.5 millions.



Document liveness detection also plays a major role in Know Your Customer (KYC), Customer Due Diligence (CDD), and Anti-Money Laundering (AML) security standards accepted by payment providers, banks, e-commerce, and various institutions worldwide.

Solutions

To address the issue, multiple solutions and countermeasures are proposed.

IDLive Doc

IDLive Doc, a solution developed by ID R&D, focuses on preventing screen replay attacks. The problem stems from the criminal tactic: perpetrators can simply obtain an image of a real ID or produce its copy in a software like Photoshop or GIMP. Then, the fabricated image is presented to the sensors of a RIDP system from a high-resolution screen.

Typically, a human eye can’t spot a forgery if the "high resolution retina displays" are used for the attack. Therefore, an accurate and quick liveness check is required. IDLive Doc is based on a passive liveness detection method, hence the procedure requires a few seconds and sets no extra verification challenges.

It is undisclosed which technical solutions are used, except that a "unique Deep Neural Network-based approach" serves as its basis. It allows detecting liveness from just a single image. As mentioned, it’s capable of performing "artifact detection", which possibly refers to editing artifacts, double compression, and other similar clues.


Smart Engines

Smart Engines introduced two solutions in the document liveness detection.

Computation Document Forensic AI

Computation Document Forensic AI (CDF AI) is a GDPR-compliant multipurpose AI system. It specializes in full-scale document analysis, providing a group of AI models for analyzing images, document liveness holograms, templates, stamps and seals, text and font, and so forth.

These AI models search for potential anomalies left by image editing, artificial synthesis, digital or physical copying, and other procedures. The procedure is based on the optical and infrared range analysis, which can be compared to a standard UV lamp check widely used in many countries.

One of the features that CDF AI also proposes is a complete autonomy. All data and analysis tools are deployed and stored on the user’s gadget — e.g. a smartphone or a tablet. In turn, this prevents data leakages and provides a convenient offline workflow.

Smart ID Engine

Smart ID Engine is an ID liveness detector that analyzes various properties of a document. Among them, authors name the document geometry, holograms and monograms, machine-readable zones (MRZ), and other security components. Besides, Smart ID Engine can attest the document state by capturing a video stream or creating separate frames.

The solution can process documents from 210 jurisdictions worldwide, while supporting 99 languages, including Chinese and Urdu. Working in a passive mode, it also shows high performance speed: it takes 250 ms per 1 frame to recognize a US driver’s license.


CheckScan

CheckScan is a novelty ID verification approach introduced in March 2022. The idea suggests that it’s possible to distinguish a legitimate document from a fake by analyzing their quality. It consists of two stages.

Feature extraction. This stage is based on Fast Fourier Transform (FFT). Basically, an image in question is separated into a group of blocks that do not overlap. Then, the FFT magnitude spectrum will be estimated for every single block. As a result, FFT magnitude peaks are extracted as discriminative features.

Hash construction. Previously obtained magnitude peaks are quantized into binary codes based on the peak coordinates. Then good discriminates can be achieved. They are capable of distinguishing bona fide and fabricated identification documents.

Dataset for Document Recognition

To assist development of the document liveness detection tools, a number of datasets exist. However, if compared to the deepfake datasets, their number is quite limited. The known examples are LRDE Identity Document Image Database (LRDE IDID), Brazilian Identity Document Dataset (BID Dataset), SmartDoc (partly), DLC-2021, and Mobile Identity Document Video Dataset (MIDV) presented in a few generations.

Virtually all those datasets either provide insufficient sample materials or feature IDs with blurred faces, precise cropping, nonexistent background, and so on. The MIDV dataset seeks to fix that, while also providing enough IDs and a variety of document types and templates, fonts, ethnicities, ornaments and other security elements. Different capture methods were also applied: office scanners and smartphones.



The ID photos were created with Generated Photos based on a StyleGAN2 approach. While the datasets preserve original templates that can be found within the European Union, their names and additional elements, like signatures, were purposely falsified.



Notably, ID photos are presented both in color and grayscale, which is a common case with authentic documents. Furthermore, a select number of samples were photographed in the "real life" ambiance — on the ground, desk, keyboard — to increase the difficulty for the recognition system.



Document Liveness Challenge

The first competition dubbed ICMV Document Liveness Challenge 2021 was held in 2021 by Smart Engines together with RAS and Laboratoire Informatique. The challenge was divided into three parts:

  1. Laminated documents & unlaminated gray copies
  2. Documents photographed with a smart gadget
  3. Unlaminated copies in color.

The DLC datasets included 1424 videos, each one captured vertically with an average length of 5 seconds. The samples were presented in the form of videos, separate frames and markups. The challenge continues to exist: the upcoming challenge will take place in November 2022, welcoming researchers in the area.

References

  1. Inside the Brussels flat where terrorists scored fake IDs
  2. Face Morphing, a Modern Threat to Border Security: Recent Advances and Open Challenges
  3. A guide to getting remote identity verification right
  4. ID.me gathers lots of data besides face scans, including locations. Scammers still have found a way around it
  5. Who is ID.me?
  6. What is Customer Due Diligence (CDD)?
  7. IDLive Doc
  8. Standalone universal ID document liveness detector launched by ID R&D
  9. General Data Protection Regulation
  10. Understanding UV Security: How and Why It’s Used on Payment Documents
  11. Smart Engines has launched document liveness detection for ID documents scanning
  12. Smart Engines has launched a new generation of recognition systems with document authentication and biometric verification
  13. CheckScan: a reference hashing for identity document quality detection
  14. LRDE Identity Document Image Database
  15. Brazilian Identity Document Dataset
  16. SmartDoc
  17. DLC-2021
  18. Mobile Identity Document Video Dataset
  19. A synthesized face created with Generated Photos
  20. MIDV-2020: A Comprehensive Benchmark Dataset for Identity Document Analysis
  21. Fall digital identity events announced by Goode Intelligence, Future Identity, EAB, Smart Engines
  22. Russian Academy of Sciences by Wikipedia
  23. The 15th International Conference on Machine Vision