High-Performance Facial Expression 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 LDA's 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.