Combating identity fraud at Europe’s borders
Identity fraud is a major threat to European security. Border guards control many people throughout the day, and don’t have time to check documents in detail. Most therefore rely on automatic verification tools. These systems can detect some types of fraud, but not all. This means: “Some fraud is ‘left aside’,” says Claude Bauzou from IDEMIA in France. There are several types of document fraud, including forgeries, counterfeits and fraudulently obtained genuine documents. One of the main types of fraud associated with border crime is a ‘morphing attack’: the fraudulent use of genuine documents that display a morphed face image. These images normally combine features from different faces, creating a composite image. In the EU-funded iMARS project, Bauzou and her colleagues developed new tools to help evaluate the authenticity and verification of ID documents, and to detect fraud. Specifically, one of the main goals was to provide solutions to detect image morphing attacks.
Developing morphing attack detection systems
The team focused on two types of morphing attack detection (MAD): single image morphing attack detection (S-MAD); and differential morphing attack detection (D-MAD). In S-MAD, the technical solution consists of one input (a potentially morphed face image of an ID document or to be printed on an ID document), and one output (a score representing the likelihood that the input image is morphed). In D-MAD, the person holds two inputs: a potentially morphed face image of an ID document and another face image corresponding to the face of the person holding the ID document. The system then delivers the score of the likelihood the first input is morphed. “In both cases, the technical solution analyses the input images, in most cases with AI-based algorithms, and computes the score,” explains Bauzou. Through iMARS, the team created many algorithms to support these detection mechanisms. They gathered data to learn and test the algorithms, which were then iterated until effective. “Different algorithm developers were crossing ideas and synchronising on data often, so that they could progress efficiently,” adds Bauzou. When using the tools, border guards will take action if they see a warning suggesting the document may have been morphed.
Raising awareness of identify fraud across Europe
The iMARS project achieved all its initial goals. “We are now reaching an accuracy level that makes the technology suitable for detecting automatically generated morphs and average-quality morphs,” says Bauzou. The project also created a MAD training programme for border guards and ID document experts, to support the verification of the algorithms. And a new software platform will measure the performance of MAD algorithms. “We have also raised awareness to European stakeholders of the risk of morphing,” adds Bauzou. “The vulnerability of current face recognition systems has been assessed to make users aware of the risks and provide improvement targets to developers.”
Continued research into morphing
The work of iMARS will continue under at least two Horizon-funded projects: EINSTEIN and SafeTravellers. On the industry side, research is also ongoing. “I see a huge need to deploy and improve the training on ID document fraud, including morphing,” notes Bauzou. “iMARS really took a huge step forward, and it would be a pity that this is not continued.”
Keywords
iMARS, fraud, identity, morphing attack, detection systems, research