This project implements a face liveness detection system using deep learning to distinguish between real faces and spoofed images.

• liveness.py: Script to run liveness detection on a single image
• model.ipynb: Jupyter notebook containing model training code

The model uses a convolutional neural network (CNN) architecture:

• Input shape: (128, 128, 3)
• 3 convolutional layers with ReLU activation
• Max pooling and dropout layers
• Fully connected layers
• Output: Binary classification (real vs spoof)

The model is trained on a custom dataset of real and spoofed face images. Data augmentation is applied during training. The training process uses:

• Binary crossentropy loss
• Adam optimizer
• 50 epochs
• Batch size of 12

To run liveness detection on a new image:

1. Ensure you have the required dependencies installed
2. Place the trained model file liveness.h5 in the same directory as liveness.py
3. Run:
4. The script will load the image, run it through the model, and output the prediction (real or spoof) along with the confidence score.

The model achieves over 90% accuracy on the training set and around 75-80% accuracy on the validation set after 50 epochs of training.

• Collect more diverse training data
• Experiment with different model architectures
• Implement real-time liveness detection on video streams

• Python 3.x
• TensorFlow 2.x
• Keras
• NumPy
• Matplotlib