Multimodal Palmprint Authentication
Show of Hands
Contactless identity from the palm of your hand. Four models study every scan and vote on the verdict, and together they reach 99.75 percent AUC.
- Python
- PyTorch
- OpenCV
- FastAPI
- Docker
- CUDA
Problem
Contactless palmprint verification has to cope with rotation, lighting, and scale variance that fingerprint sensors never see — a single model tends to overfit to one invariance and miss others.
Approach
- Four-way ensemble combining CNN backbones with handcrafted texture features, so learned and engineered representations cover for each other’s blind spots.
- Late-fusion soft voting over the four scorers’ outputs for the final verification decision.
- CUDA-accelerated PyTorch training; FastAPI serving layer; Docker for reproducibility.
Result
99.75% AUC on verification — strong evidence that ensemble diversity (deep + handcrafted) beats any single track on biometric robustness.