Facial Expression Recognition Based on Quaternion-Space and Multi-features Fusion

摘要

There is an increasing trend of using feature fusion technique in facial expression recognition. However, when traditional serial or parallel feature fusion methods are used, the problem of highly dimensional features and insufficient fusion of possible feature categories always exist. In order to solve these problems, a novel facial expression recognition method based on quaternion-space and multi-features fusion is proposed. Firstly, four different kinds of expression features are extracted such as Gabor wavelet, LBP, LPQ and DCT features, then PCA+CCA framework is proposed and used to reduce the dimensions of the four original features. Secondly, quaternion is used to construct the combinative features. Thirdly, a novel quaternion-space HDA method is proposed and used as the dimensional reduction method of the combinative features. Finally, SVM is used and set as the classifier. Experimental results indicate that the proposed method is capable of fusing four kinds of features more effectively while it achieves higher recognition rates than the traditional feature fusion methods.