The main goal of this project is to check whether the given image is a real or fake face (liveness) and compare between 2 faces which is expected to have a high of accuracy and light weight. This repo is inspired by Silent Face Anti Spoofing and uses some of the file and code in it, but we use our own custom model.

• Automatically align facial landmarks before feeding them to the model for testing.

• Output after model testing, returns "Real" or "Fake" with a confidence score.

conda env create -f environment.yml

Activate venv:

conda activate python-ekyc

Tested on python 3.8

pip install -r requirements.txt

Place your image in sample folder.

Command to run:

python main.py --filename suzuka-nakamoto.jpg

Output:

Face detected
Image 'suzuka-nakamoto.jpg' is Real Face.
Score: 1.39.
Blur score: 345.25
Elapsed time 2.80 s

• Auto rotate image
• Align face
• Detect spoof image/face
• Compare between 2 faces

• The images provided must have a 3:4 (w:h) ratio.
• If the face is rotated, e.g 90 degrees, the depiction of facial landmarks is invalid.

• davisking/dlib
• minivision-ai/Silent-Face-Anti-Spoofing