simple app to register and recognise face using DeepFace

It comes with the two simple features: Register and Recognize. In the 'Register Face' option, one may click his/her photo from the web/front camera or may select an image from the gallery and enter the name of the person. This photo is analyzed by DeepFace, and a cropped image is stored along with the entered name. Now, when the user submits similar images of the registered person, the face is recognized, and the name is displayed.

git clone https://github.com/s0ubhik/facecog
cd facecog

python3 -m venv venv
source venv/bin/activate

pip install -r requirements.txt

flask run -p 8080

Open the following link in the browser

http://127.0.0.1:8080

docker build -t facecog .

Now that our image is built we need to start the image, note that in the following command -p paramater is used to tunnel a port from the container to your machine 8080:5000 means the 5000 port of the image is tunneled to 8080 port of your computer

docker run -p 8080:5000 -i facecog

.
├── app.py
├── faces.db    ( Names along with hashes )
├── dataset/    ( Image dataset )
├── static/
└── templates/

When a user hits register a HTTP POST request is sent to /api/register that contains two field the name of the user and the base64 url-encoded image.

On the server-side the image is passed to DeepFace.extarct_faces which retruns the array of cropped faces found, now we check for few condition where is there is a face or there are multiple faces.

After this we generate a md5 hash of the cropped face, and sa store the cropped face as jpg in the dataset/ folder i.e dataset/<hash>.jpg, now store the name of the user along with this hash in the tables faces of faces.db

When a user hits recognise a HTTP POST request is sent to /api/recognise with the base64 url-encoded image of the user.

On the server-side the image is passed to DeepFace.find along with the directory of dataset, this method returns as arraw of matched faces along with their filepath and a value that represents the score of mathc, out of them we only that the first entry with best score.

Now we extract the hash from the file path and fetch the corresponding user.

Distributed under the MIT License. See LICENSE for more information.