Génération d'image faciale synthétique à base de points caractéristiques.

Abstract

Avatars and cartoons are widely used in various areas of emerging technology, from augmented reality to social robots. These avatars help to perceive information in a natural way and play a key role in transmitting various emotional signals. The aim of our project is to design an intelligent system capable of transferring the characteristics of a real face to a cartoon face, enabling this avatar to be used in a variety of contexts. Our approach is based on recent advances in deep learning, in particular adversarial generative networks and autoencoders. The proposed system starts by extracting minutiae from each image, which are then used as input to the autoencoder. The minutiae extraction module is in fact the encoder of the autoencoder, which has been trained on the FERG-DB database and has achieved 78% accuracy. The output of this module is then used as input for the cartoon generation module. To carry out this generation, we used the StyleGAN3 generator, previously trained by knowledge transfer from the FERG-DB database. The use of autoencoders and the StyleGAN3 generator represents an interesting approach to achieving the objective of this project, and the results obtained so far are encouraging, with signi cant potential for improvement through more advanced training.