Abhishek12233/fingerprint-embedding-resnet18

Fingerprint Embedding ResNet18

A prototype fingerprint embedding model based on pretrained ResNet18.

Model Details

• —Architecture: ResNet18
• —Input Size: 224 × 224
• —Input: Grayscale fingerprint converted to 3 channels
• —Embedding Size: 512
• —Output: L2-normalized embedding
• —Similarity Metric: Cosine Similarity

Current Test Results

• —Same Fingerprint Similarity: 0.9128
• —Different Fingerprint Similarity: 0.8878

Pipeline

Fingerprint Image → Preprocessing → ResNet18 → 512-D Embedding → L2 Normalization → Cosine Similarity

Status

This model is currently a prototype.

The ResNet18 backbone has not yet been trained specifically for fingerprint recognition.

Future Work

• —Train on a fingerprint dataset
• —Siamese Network
• —Contrastive / Triplet Loss
• —Improve embedding separation
• —FAR / FRR / EER evaluation
• —Export to ONNX

Warning

Do not use this prototype for production biometric authentication.