Facephi Engineers Publish New Scientific Research on Document Fraud Detection in IEEE

Researchers from Facephi’s R&D Department have published a new scientific paper in IEEE Transactions on Biometrics, Behavior, and Identity Science, one of the leading international journals in the field of biometrics. The study presents a Few-Shot Learning-based solution capable of detecting forged identity documents in countries where very limited data is available to train artificial intelligence models.

The paper, titled “From Simulation to Production: Few-Shot Learning for ID Card Presentation Attack Detection”, was developed by Álvaro S. Rocamora and Juan M. Espín, professionals from Facephi’s R&D department, together with Juan E. Tapia, who is affiliated with the Biometrics and Internet Security research group at Hochschule Darmstadt (h_da) in Germany.

This new publication adds to the ongoing scientific activity promoted by Facephi among its engineers and once again highlights one of the key challenges in digital identity: how to protect verification processes against document fraud even when the data available for training AI systems is limited.

The Problem This Research Addresses

When someone attempts to impersonate a digital identity, one of the most common techniques is presenting a fake document: a photocopy, a printed copy, or even a screen displaying an image of an ID card or passport instead of the real physical document. Detecting these attempts, known as presentation attacks, is one of the key functions of any digital identity verification system.

The challenge is that these systems need to learn the patterns of each document type and each country, and traditionally this has required thousands of real-world examples. When a company wants to deploy its solutions in a new market, that amount of data is rarely available at the beginning.

The team has demonstrated that this challenge can be addressed through a technique known as Few-Shot Learning, which enables a model to learn how to identify fake documents or screen-based attacks from a very small number of samples—in this case, just 50—instead of requiring large volumes of training data.

To validate this methodology, the research focused on a specific case study: starting with a system initially trained on data from Chile, the team tested its ability to extend fraud detection capabilities to Nicaragua, Honduras, and El Salvador, countries with significantly less available data, while maintaining a very high level of accuracy. These results confirm that this approach can adapt to any country or market where data availability is limited, opening the door to broader applications in the future.

“One of the major challenges in biometric research is developing solutions that can generalize when the available data is extremely limited, something that is particularly important in a field as sensitive as digital identity. Through this work, we demonstrate that it is possible to adapt models to new countries using very few examples, enabling the deployment of document fraud detection systems in markets where data is scarce. We believe that this type of research contributes to addressing a challenge that remains unresolved due to the high variability between documents, countries, and types of attacks.” Álvaro S. Rocamora, R&D Senior AI Researcher at Facephi.

Applied Research: A Key Tool Against Identity Fraud

Identity fraud is one of the most significant challenges facing digital verification today, and the lack of data for certain countries or document types remains one of the main technical barriers to combating it effectively.

Researching new ways to train artificial intelligence models with less data is not only an interesting technical challenge—it is a real industry need, especially in a context where access to and use of sensitive data are increasingly regulated.

Publishing this type of advancement in peer-reviewed scientific journals, as in this case, is also a way to ensure the rigor of the work and share it with the international research community working on the same challenges.

A Team Effort, Both Within and Beyond Facephi

This research was led by Álvaro S. Rocamora and Juan M. Espín from Facephi’s R&D department, together with Juan E. Tapia from the Biometrics and Internet Security research group at Hochschule Darmstadt (h_da), one of Europe’s leading institutions in biometric research.

“This work highlights the strong relationship between Facephi and leading scientific institutions, bringing a research project to completion with a publication in a prestigious journal. At Facephi’s Research Department, we strongly support these collaborations as a way to improve, evolve, and build new products.” Juan Manuel Espín, R&D PAD Lead.

The work has also received support from the National Research Center for Applied Cybersecurity ATHENE, as well as funding from the German Federal Ministry of Education and Research and the Hessian Ministry of Higher Education, Research, Science and the Arts.

This collaboration between the Facephi team and international academic research centers is another example of the company’s commitment to applied research, and of how Facephi’s internal talent actively contributes to scientific projects alongside leading European institutions in the field of biometrics.