Biometric matching is a process of verifying a person’s identity based on unique biological characteristics, that include face, voice, fingerprints, and others. It is used in various applications such as security systems, access control, and identity verification, and currently is an important part of security systems.
Inspired by the multimodality of human perception and the ability to make decisions based on any combination of available data, we introduce the first input-agnostic, multimodal, and unified biometric matching system.
Our approach is designed to handle unimodal, multimodal, missing- modality, and cross-modal scenarios, supporting verification using audio, visual, and thermal modalities:
Link to the paper: One Model to Rule Them All: A Universal Transformer for Biometric Matching
Link to the Github: https://github.com/IS2AI/unified_multimodal_transformer
If you use our project in your work, we kindly ask you to cite our paper:
M. Abdrakhmanova, A. Yermekova, Y. Barko, V. Ryspayev, M. Jumadildayev and H. A. Varol, “One Model to Rule Them all: A Universal Transformer for Biometric Matching,” in IEEE Access, vol. 12, pp. 96729-96739, 2024, doi: 10.1109/ACCESS.2024.3426602. keywords: {Transformers;Vectors;Feature extraction;Visualization;Speech recognition;Biological system modeling;Task analysis;Biometrics (access control);Biometric matching;cross-modal matching;face verification;face-audio association;metric learning;multimodal verification;speaker verification;transformer},
