Face recognition with eye detection

1. Install python3

On MacOs  
brew install python3  

On Ubuntu  
sudo apt-get install python3  

2. Setup a virtual environment in the folder sudo pip3 install virtualenv
virtualenv myenv
source myenv/bin/activate

3. Install requirements pip3 install -r requirements.txt
   ├── pandas
   ├── numpy
   ├── sklearn
   ├── flask
   ├── flask_cors
   ├── imutils
   ├── cmake
   ├── dlib
   ├── opencv-python

├── dataset
│   └── harshith
├── detect_blinks.py
├── extract_embeddings.py
├── face_detection_model
│   ├── deploy.prototxt
│   └── res10_300x300_ssd_iter_140000.caffemodel
├── facerecog_v2.py
├── face_registration.py
├── flask_backend.py
├── haarcascade_frontalface_default.xml
├── openface_nn4.small2.v1.t7
├── output
│   ├── embeddings.pickle
│   ├── le.pickle
│   └── recognizer.pickle
├── pycache
│   └── flask.cpython-37.pyc
├── README.md
├── recognize.py
├── recognize_video.py
├── requirements.txt
├── shape_predictor_68_face_landmarks.dat
├── test
│   └── test.png
└── train_model.py

• Face registration (collecting images)
• Face recognition with blink detection

Dataset: Labelled faces in the wild Download

1. For using a custom dataset, run python3 face_registration.py --name [person_name] to collect images of a person in folder dataset/[person_name]
2. Extract embeddings from an image python3 extract_embeddings.py. The embeddings are stored in the path output/embeddings.pickle

Added flask server to make an api call
python3 facerecog_v2.py

• Using face alignment [2] to get the facial landmarks and using [1] to compute eye aspect ratio to identify eye blinking.
• This is to avoid spoofing attacks by clearly differentiating a real person's face and an image of the person

• Generate 128-d embedding of each person's image in the dataset.
• Train the SVM model for classification
• Testing: Generate the embedding of the person you want to recognize and compute the similarity between test person's embedding and pre-computed embedding.

1. Real time eye detection
2. Face alignment
3. Shape predictor dataset
4. FaceNet
5. dlib pretrained models