Manifold learning approach to facial expression recognition on local binary pattern features

In this paper, the facial expression recognition (FER) is investigated based on the observation that a sequence of images of a certain facial expression define a smooth manifold. First, local binary pattern (LBP) algorithm is used to extract the local texture features of the expression images. Then, locally linear embedding (LLE) method is used to learn the structure of the expression manifold in the LBP feature speace. Finally support vector machine (SVM) is used for the classification of expressions. The LBP+LLE algorithm is experimented on the Japanese female facial expression (JAFFE) database. Extensive experiment result comparisons show that LBP features and manifold approach are effective methods for FER. Their combination provides much better performance compared with that of those traditional algorithms such as PCA, LDA, etc.