Ziyu Ding MSc Dissertation 2014/15

Emotion Recognition System Based on Facial Expression Features

Supervised by Y.Gotoh

Abstract

As huamn-computer interaction becomes a significant field, the emotion recognition system gets more attention. Since the emotion recognition system requires machine to be able to identify different human emotions, applications of the system are variours such as facial expression recognition apps and security protection apps. To achieve a better emotion recognition system, many researchers focus on this area and try to improve its accuracy and efficiency of it. Although several methods are provided to recognise human emotions, in this report, the chosen implementation method is based on human facial expression to classify seven emotions: anger, disgust, hapiness, sad, surprise and neutral. The techniques that are used to build the system are Local Binary Pattern (LBP) and Kernel Principle Component Analysis (KPCA) for facial feature extraction and Support Vector Machine (SVM) for training and classification. The syetem training and testing are based on Japanese Female Facial Expression (JAFFE) open source database while evaluation and testing are based on both JAFFE database and Yale Faces database. The results of the emotion recognition system proved the system is successfully based on JAFFE database, however, it also indicates the improvement possibility for Yale Faces database and so on.