Upper Facial Action Units Recognition Based on KPCA and SVM

Abstract

The existing methods of facial action unit recognition are always affected by illumination and individual difference. An upper facial action units recognition method based on KPCA and SVM is presented in this paper. In this method, KPCA algorithm which is formed by choosing and designing kernel function in terms of visual features of upper facial action units is used to extract the upper facial action unit feature and the two features associated with illumination effect are removed. Then the optimal kernel function and chastisement factor in SVM algorithm are determined by experiments. Finally the SVM is used to classify and recognize action units. This method is tested on the Cohn-Kanade 's facial expression image database. The average recognition rate achieves 90.6% and the recognition speed is also fast. The experiments show that this method is not sensitive to illumination and individual difference and can be used to real time recognition.