Emotion recognition through facial expression in human-computer interaction

• 主要作者: Loh, Chor Peng.
• 其他作者: Yap Kim Hui
• 格式: Final Year Project
• 語言: English
• 出版: 2010
• 主題: DRNTU::Engineering::Electrical and electronic engineering::Electronic systems::Signal processing
• 在線閱讀: http://hdl.handle.net/10356/40294

Humans are able to detect and interpret faces and facial expressions during interactions with little or no effort required. A system that could perform this accurately would form a big step in improving the human-computer interaction between man and machine. Development of such a system to interpret faces and facial expressions requires the following three major stages: detection of face segment from an image, feature extraction and classification of facial expressions. This project is concern about performing emotions recognition on static facial images through the development of a facial expression recognition system. In this study, a facial normalization algorithm was implemented and carried out on the static images based on the detection of eyes before performing feature extraction through Eigenface and Fisherface algorithms. Next, we conduct experiments on the effect and relevance of facial normalization when presenting the extracted feature vectors to the classifiers, Neural Network and Support Vector Machines (SVMs) for performance evaluation. The results of our evaluation had shown that our facial expression recognition system performed consistently better on facial image that are been pre-processed with normalization when compared to images that are not been normalized. In the generalization classification on unseen, novel individuals, our normalized techniques achieved 61% accuracy when presented to the Eigenface algorithm with linear based One Against All (OAO) SVM as compared to 54% without normalization using the same feature extraction method and classifier. Similar evaluation of normalized unseen, novel images using Eigenface with Radial Basis Function (RBF) based OAOSVM yield an accuracy of 64%.