Résumé:
In recent years, synthetic face generation technology has developed rapidly, based on
deepfake technology to create ultra-realistic fake images and videos.
In this master thesis, our goal is to develop a fake face detection system using a specific
neural network called Visual Transformer (VIT).
This deep learning approach allows us to obtain a simplified representation of our data,
on which we distinguish "fake" images from "real" images.
Originally used in natural language processing where they proved their robustness and
accuracy, VITs were later adopted in various areas of image processing and machine vision.
The proposed system consists in detecting the false information received on images or
videos through the detection of false faces detected.
Testing of falsified faces has yielded encouraging results, but improvements can be
made by continuing to learn.
