FingerPrintMatchingSNN

A project implementing fingerprint matching using Siamese Neural Networks (SNNs) for efficient and accurate pattern recognition.

This repository explores advanced biometric authentication through SNNs, leveraging similarity-based distance approach for real-time fingerprint analysis.

• Efficient: Utilizes lightweight CNN subnets to capture features and calculate the similarity using L2 distance metrics.
• Accurate: Achieves high matching precision on standard datasets.
• Scalable: Modular design for integration into larger biometric systems.

➤ Table of Contents

• Installation
• Getting Started
• Usage

➤ Installation

1. Clone the repo

   git clone https://github.com/OmGhag/FingerPrintMatchingSNN.git

2. Create a Python virtual environment (Python 3.8+ required)

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies

   pip install -r requirements.txt

   Key libraries include: NumPy, Keras, Tensorflow, OpenCV (for image processing).

➤ Getting Started

To get started quickly, follow these steps in order:

1. Run the data preparation script to set up the dataset:

   python data.py

2. Open and run all cells in preprocess.ipynb to preprocess the fingerprint data (e.g., minutiae extraction and feature engineering).
3. Finally, open and run all cells in train.ipynb to train the SNN model on the preprocessed data.

➤ Usage

Data Preparation

Run the script to load and organize the fingerprint dataset:

python data.py

This handles downloading or loading sample data into the appropriate directories.

Preprocessing

Use the Jupyter notebook for data cleaning and feature extraction:

• Launch Jupyter: jupyter notebook preprocess.ipynb
• Execute all cells sequentially. This includes steps like image enhancement, minutiae detection using OpenCV, and normalization for SNN input.

Training the Model

Train the SNN using the dedicated notebook:

• Launch Jupyter: jupyter notebook train.ipynb
• Execute all cells. This covers model architecture setup (Leaky Integrate-and-Fire neurons with STDP), training loop, and evaluation.

Key Features

• Minutiae Extraction: Ridge endings and bifurcations detected via OpenCV.
• SNN Architecture: Dual CNN Subnet architecture making it lightweight and scalable.
• Evaluation Metrics: .

After training, saved models can be used for inference in custom scripts.