Facial expression recognition with enhanced feature extraction using graph fourier transform

A novel method for the facial expression recognition is proposed using the graph Fourier transform. In addition, the dimension of the feature vector is reduced by extracting the selective frequency components. The interesting observation is that the few eigenvectors corresponding to lower frequencies provide comparative accuracy in classification as using all the frequencies. Here, the spectral analysis of the facial graph signals suggests that the interrelationship of the graph signals is better captured in the lower frequencies. Experimental results on CK+ and JAFFE datasets demonstrate the effectiveness of the proposed method.