Etudiant : ZABOUL HAMZA

Filière : LF Sciences Mathématiques et Informatique

Encadrant : Pr. SATORI HASSAN

Annèe : 2022

Résumé : This project developed research to classify facial emotions over facial and reel time images using deep learning techniques. This is a complex problem that has already been approached several times with different techniques. While feature engineering achieved good results, this project focused on feature learning, one of DL’s promises. While the results achieved were not state-of-the-art, they were slightly better than other techniques, including feature engineering. It means that eventually, DL techniques will be able to solve this problem given enough amount of labelled examples. In contrast, feature engineering is not necessary. Image pre-processing boosts classification accuracy. Hence, it reduces noise in the input data. We also discussed collecting the database and merging it with external ones. The goal was to reach more than 60% accuracy and identify seven emotions in our database. We have achieved 70% of a target due to our ability to reach 90% recognition accuracy and 65% in validation accuracy, and we were able to recognize seven emotions . We could not achieve a better accuracy but due to the problems that we face in every step of the project: • one of the main problems is people refusing to take pictures due to there insecurity and specially girls. • a lot of people cannot express all the emotions (almost all of them can express just happiness and surprised and neutral) • lack of good equipement because training the model need a good GPU and CPU so it was very difficult to train model with just our laptops.