Facial Expression Recognition using KCCA with Combining Correlation Kernels and Kansei Information

Abstract

Kernel canonical correlation analysis (kCCA) with combining correlation kernels of multiple-orders and Kansei information is applied to facial expression recognition. Any explicit feature extraction is done and spatial correlation features of image data are implicitly incorporated in the correlation kernels. Further, Kansei information is included as the second feature in kCCA. Classification experiments with JAFFE database show that, although the use of Kansei information in itself gives lower classification performance than class indicators optimal for classification tasks, combining Kansei information with them makes the classification performance higher than the only use of the indicators.