Challenge: Dual Factor Lip-Based Biometric User Authentication

About the Challenge

This competition focuses on dual factor biometric authentication using spoken passphrases and lip motion from video data. Participants must develop systems to authenticate users in an open-set scenario with high accuracy and low error rates.

• Spoken passphrase + lip motion
• Open-set authentication
• Evaluated by Equal Error Rate (EER)

Evaluation

Primary metric: Equal Error Rate (EER)

Participants must submit a JSON file like:

{
  "video_file": "sample_01.mp4",
  "score": 0.87,
  "user_id": "U001",
  "spoken_word": "open"
}

Dataset Structure

Each user has a folder with multiple videos containing correct and incorrect passphrases. The system must identify valid users and reject impostors or invalid phrases.

Example of the Training Set filename format: <ID>_<NUM_PROG>_<T/F>.mp4, where ID is the user identifier, NUM_PROG is the progressive number, and T/F indicates whether the video contains a correct (T) or incorrect (F) passphrase.

Research Topics

Rules

• Participants can receive the training set and its annotations after team registration
• The test set will be released after the submission of 1 page abstract about the method.
• Particpant will submit the test results (in JSON) format within 5 days after test set release.
• The participants are strongly encouraged to submit a paper by the deadline of 20th June, 2025 (extended). The paper can be submitted through OpenReview (link soon available). Authors can find complete instructions of how to format their papers here. The maximum number of pages is 12 including references. Accepted papers will be included in the ICIAP 2025 Workshops Proceedings. Camera-ready submission: July 10th, 2025 (extended)

Organizers

• Massimo Orazio Spata – University of Catania
• Georgia Fargetta – University of Catania
• Alessandro Ortis – University of Catania

Bibliography

• Massimo Orazio Spata, Alessandro Ortis, Sebastiano Battiato, CNNMC: A Convolutional Neural Network with Monte Carlo Dropout for Speaker Recognition, 25 May 2025, PREPRINT (Version 1) available at Research Square, https://doi.org/10.21203/rs.3.rs-6570040/v1

Sponsors