Facial Expression Recognition using Hybrid CNN-SVM Technique

Abstract

Analysis of facial expression rapidly becomes an area of intense interest in computer science and the design of communities for human-computer interaction. The most descriptive manner in which humans show emotions is through facial expressions. Despite efforts to improve different facial expression recognition methods, well-designed features are still open issues that affect classifier performance. This paper designed a facial expression recognition system that used convolutional neural network architecture with fully connected layers in order to extract the facial features then classify the extracted features based on the support vector machine classifier. The facial expressions are classified into one of seven expressions (neutral, happy, sad, fear, surprise, anger, disgust). The experiment is conducted on the Japanese Female Facial Expression (JAFFE) to calculate the accuracy of the proposed system. The system achieves a 95.25 percent accuracy for recognition of facial expression from the image on the grayscale. The results achieved show that the system provides improved classification accuracy when compared to other systems in the literature.

Keywords: Facial expression recognition (FER), Convolution neural network (CNN), Facial expression (FE), Support vector machine (SVM).