ID card PAD competition at IJCB 2025 builds on lessons from inaugural event

The winning entries to a competition for technologies to detection presentation attacks using ID cards will be presented during the International Joint Conference of Biometrics (IJCB) 2025, on September 8 to 11 in Osaka, Japan.

Organized by Facephi, Fraunhopher-IGD and Hochschule Darmstadt (h-da), PAD-ID Card 2025 is the second edition of the research competition in support of remote biometric verification systems. PAD is typically associated with the probe image used in the biometric matching step of the ID verification process, but guarding against spoofed reference images is necessary for the subsequent step to have any value.

The competition to be held during IJCB 2025 follows up on the inaugural edition at least year’s event.

The second competition offers an independent assessment of ID card PAD algorithms with print, screen and composite attacks based on a PVC card template, an approach the organizers say is aligned with current state of the art open datasets. A second track provides an evaluation protocol that includes attacks and bona fide ID card images from volunteers’ IDs for use by researchers, and is aligned with approaches using commercial and proprietary datasets. Both tracks are being held for research purposes only.

Registration is now open to companies and universities through the PAD-ID Card 2025 website.

Results from 2024 competition

The first PAD-ID Card competition featured eight entries from five teams, plus three baseline systems. Two of the teams hailed from NTNU. The models they developed for detecting fake IDs were tested against datasets of IDs taken from four countries.

They were judged in terms of AVrank, which weights bona fide presentation classification error rates (BPCERs) for different attack presentation classification error rates (APCERs).

The competition was one by an anonymous team, which scored 74.8 percent AVrank with one of its two submitted models. Team IDVC, from the ID Vision Center came second with 77.65 percent for one of its two models submitted.

The organizers concluded that the “results show that the generalisation capabilities to predict PAD between different countries and attacks are still challenging.” This is partly due to limitations in the open datasets available for algorithm training, which tend to present fewer bona fide images per subject and generate many attacks from each ID card, resulting in an imbalance.

The results also show that ID cards based on ICAO standards are easier to detect and accurately classify.

Article Topics

biometrics  |  digital ID  |  document liveness  |  document verification  |  FacePhi  |  Fraunhofer Institute IGD  |  identity document  |  IEEE  |  IJCB  |  PAD-ID  |  PAD-ID Card 2025  |  presentation attack detection