Skip to Main Content
The paper presents a facial expression recognition approach based on wavelet transform features and neural network ensemble classifier for the six basic facial expressions from static images of the CMU-PITTSBURGH AU-Coded Face Expression Image Database. Because the facial expression information are mostly concentrate on facial expression information regions, so the mouth, eye and eyebrow regions are segmented from the facial expression images firstly. Then the low-dimension features using wavelet transform and Karhunen-Loeve transform are acquired. A neural network ensemble classifier based on Bagging is constructed finally. The proposed approach has demonstrated superior performance compared to neural networks. The neural network ensemble based classifier yielded an accuracy of 98.5%; the best accuracy obtained from all other neural network based classification schemes tested using the same database.