A production-ready biometric authentication system built with Flask and TensorFlow. It uses a Siamese neural network to compute fingerprint similarity, combining advanced preprocessing (Rembg, OpenCV, CLAHE, Otsu thresholding) with a web-based interface.
This project enables secure fingerprint verification by comparing uploaded images against stored templates using deep learning. The core architecture leverages a Siamese network trained to measure similarity via Euclidean distance.
- 📦 Backend: Flask + TensorFlow
- 🧠 Model: Siamese Network (contrastive loss)
- 🎛️ Preprocessing: Rembg, OpenCV, fingerprint-enhancer
- 🌐 Interface: HTML + JS + REST API
- 📊 Output: Similarity score + binary match decision
Fingerprint-sensor/ ├── server.py # Main Flask application ├── verify.html # UI interface (HTML/JS) ├── train_siamese.py # Model training script (optional) ├── siamese_fingerprint_model_v2.keras # Trained model ├── stored_fingerprints/ # Stored template fingerprints ├── uploads/ # Temporary uploaded images ├── processed/ # Processed fingerprint masks └── README.md
- Siamese CNN with shared weights
- Input shape:
(128, 128, 1) - Distance metric: Euclidean
- Threshold-based classification (e.g.,
distance < 0.3 → match) - Custom loss:
ContrastiveLoss(margin=1.0)
bash git clone https://github.com/Shark-Bot-X/Fingerprint-verification-system cd fingerprint-verification-app 2. Install Dependencies pip install -r requirements.txt
Dependencies include:
flaskflask-cors
tensorflow
opencv-python
rembg
numpy
scikit-image
fingerprint-enhancer
⚙️ API Endpoint POST /verify Payload: multipart/form-data
Field Type Description name string Identifier of the stored template file file Fingerprint image (.jpg/.png)
Response: { "similarity": 0.276, "match": true, "processed_mask": "/processed/processed_input_mask.jpg" } 🔍 Fingerprint Preprocessing Pipeline Background Removal → rembg
CLAHE Normalization → Adaptive histogram equalization
Enhancement → fingerprint-enhancer
Binarization → Otsu thresholding
Resize + Normalize → Final (128, 128, 1) tensor
📈 Model Training (Optional) To train your own model on custom fingerprint pairs, use:
python train_siamese.py Customize:
Dataset structure
Augmentation pipeline
Threshold tuning for optimal accuracy
📸 Sample UI
📌 Use Cases Identity verification systems
Access control/authentication terminals
Biometric research experiments
Forensics & digital evidence validation
📄 License This project is licensed under the MIT License. Feel free to use, modify, and distribute with attribution.