Liveness Detection Demo

This is a real-time face liveness detection demo using:

• OpenCV DNN Face Detector (Caffe SSD)
• Keras/TensorFlow liveness classifier
• imutils threaded webcam capture

The script opens your webcam, detects faces, and classifies them as real or fake.

---

1. Requirements

• Python 3.8–3.11
• Webcam
• Python packages:
  • opencv-python
  • imutils
  • tensorflow (includes Keras)
  • numpy

Install dependencies:

# Create and activate a virtual environment (recommended)
python -m venv .venv
# Windows: .venv\Scripts\activate
# macOS/Linux:
source .venv/bin/activate

pip install --upgrade pip
pip install opencv-python imutils tensorflow numpy

---

2. File Structure

.
├── liveness_demo.py                 # The demo script
├── liveness.model                   # Pre-trained Keras model
├── le.pickle                        # LabelEncoder pickle file
└── face_detector/                   # OpenCV Caffe SSD face detector
    ├── deploy.prototxt
    └── res10_300x300_ssd_iter_140000.caffemodel

Download the ResNet-10 SSD face detector model files (deploy.prototxt and res10_300x300_ssd_iter_140000.caffemodel) and place them in the face_detector/ directory.

---

3. Usage

python liveness_demo.py   --model liveness.model   --le le.pickle   --detector face_detector   --confidence 0.5

Arguments:

• --model: Path to the trained Keras model
• --le: Path to the label encoder pickle file
• --detector: Path to face detector folder
• --confidence: Minimum probability to filter weak detections (default: 0.5)

Controls:

• Press q to quit

---

4. Expected Output

A window titled "Frame" will display your webcam feed. Detected faces will be highlighted with a red bounding box and labeled with the prediction:

real: 0.9876

or

fake: 0.8431

---

5. Notes

• liveness.model and le.pickle must match the preprocessing in the script:
  • Face resized to 32×32
  • Normalized to [0, 1]
• If your model uses different input sizes or preprocessing, update:

  face = cv2.resize(face, (32, 32))
  face = face.astype("float") / 255.0

---

6. Troubleshooting

• Black window / No camera: Check webcam connection or change src=0 to src=1
• Missing files: Ensure deploy.prototxt and .caffemodel are in face_detector/
• Shape mismatch: Adjust preprocessing to match model training
• Label index error: Ensure label encoder matches model’s output classes

---

7. Disclaimer

This script is for educational purposes only and should not be used as the sole factor in real-world authentication or identity verification.

---

8. Quick Start

python -m venv .venv
source .venv/bin/activate
pip install opencv-python imutils tensorflow numpy

mkdir face_detector
# Place deploy.prototxt and res10_300x300_ssd_iter_140000.caffemodel in face_detector/

python liveness_demo.py   --model liveness.model   --le le.pickle   --detector face_detector