Performance evaluation of different support vector machine kernels for face emotion recognition

Face emotion recognition systems identify emotions expressed on the face without necessarily identifying the person involved, as in Face recognition. Support Vector Machine (SVM) has been shown to give better performance on other classification tasks but has not been applied to emotion recognition, especially with still face images. This research work analyses the performance of four different SVM kernels (Radial Basis Function, Linear Function, Quadratic Function and Polynomial Function) for face emotion recognition. A database of 714 face emotion images was created by capturing twice, seven facial expressions of 51 persons with a digital camera. Principal component analysis was used to extract distinctive features by reducing the dimensionality of each image from 571 × 800 pixels to four smaller dimensions; 50 × 50, 100 × 100, 150 × 150 and 200 × 200 pixels. The performance of four SVM kernels were evaluated for face emotion recognition with 476 training and 238 testing to recognise seven emotions; Fear, Anger, Disgust, Happiness, Sadness, Surprise and Neutral. The SVM multi-class classification scheme was used in the design of our experiments. Empirical results indicate that the Quadratic Function SVM kernel performs best for face emotion recognition with an average accuracy of 99.33%. Also, larger dimensions of the reduced image results in better performance accuracy though with increasing computation time. We intend to experiment on other classifiers for emotion recognition in our future work.