Facial Action Units Recognition Based on Fuzzy Kernel Clustering

Facial action units (AUs) recognition is a challenging problem with many applications and is rapidly becoming an area of intense interest in research field of machine vision. Nonadditive AU combinations in which the appearance of the constituent AUs does change greatly increase the difficulties of AU recognition. Most AUs recognition methods treated each AU combination as a new AU. However, these methods are impractical because there are over 7000 different AU combinations. In this paper, a novel approach based on fuzzy kernel clustering is proposed to recognize AUs and AU combinations. Instead of treating each AU combination as a new AU, the proposed approach tackles AU combinations recognition by allowing each data point belongs to each AU class to a degree specified by a membership grade. The experiments show that the proposed approach is computationally simple, easy to implement. Furthermore, Experimental results indicate that the proposed algorithm generates higher accuracy for the Cohn-Kanade database and achieves better performance than the previous AUs recognition based on ANN.