Summary: The study proposes a multi-stage interpretable fingerprint matching network, IFViT, which employs a Vision Transformer (ViT)-based Siamese Network for capturing long-range dependencies and the global context in fingerprint pairs. This approach provides dense pixel-wise correspondences of feature points, allowing for an enhanced level of interpretability in the matching stages and fulfills the need for both local and global representation analysis.

Key Points:

• Vision Transformer used in a layeuble dense registration module for enhanced interpretability.
• Fixed-length representation extraction through retrained ViTs for accurate matching.
• Eccellent performance on various fingerprint database benchmarks.

Opinion: The innovative use of ViTs in interpreting and aligning fingerprints marks a significant direction in the field of biometric security. IFViT’s ability to provide a comprehensive analysis of both local and global patterns in fingerprints highlights its potential in both academic and practical applications.