High- Performance Facial Ezpression Recognition Using Gabor Filter and Probabilistic Neural Network
This paper presents a new person-independent facial expression recognition method based on Gabor filter bank, Linear Discriminate Analysis (LDA) and probabilistic neural network (PNN). At first preprocessing is performed, and then the Gabor filter bank and LDA algorithm are applied on the images. Since there are fewer image samples compared to their dimensions, a combination of principle component analysis (PCA) and LDA is used to increase LDAs efficiency. Finally the images are categorized into 6 different forms of basic emotions including happiness, sadness, anger, surprise, fear and disgust using a probabilistic neural network that is faster than other neural networks. The Cohn-Kanade database is used to train and evaluate the algorithm. The results of the test on this database reveal that the proposed algorithm has a high average performance of about 89％ in person independent facial expression recognition.
Facial ezpression recognition Gabor filter bank Linear Discriminate Analysis (LDA) Principle Component Analysis (PCA) Probabilistic Neural Network (PNN)
Saeid Fazli Reza Afrouzian Hadi Seyedarabi
Electrical Engineering Department,Zanjan University,Zanjan,Iran Faculty of Electrical Engineering,University of Tabriz,Tabriz.Iran
国际会议
上海
英文
2622-2625
2009-11-20（万方平台首次上网日期，不代表论文的发表时间）