Automatic Facial Expression Recognition using Shallow Convolutional Neural Network

Abstract

In the current decays, facial expression recognition is the important and active research area in the field of computer vision. Many researchers utilized various hand crafted features and deep convolutional neural network (DCNN) methods to improve the performance of the existing system. However, the current hand crafted system performs well on datasets that are captured in predefine conditions, and deep CNN performs well where dataset contains large amount of data. In small dataset such as CK+ and JAFFE the DCNN going to be overfitted and does not perform well. To solve these problems, in this paper, an end to end shallow CNN (SCNN) architecture is proposed. The proposed SCNN automatically organizes several characteristics of individual facial expression to support quick, accurate and reliable identification and recognition of expression of a facial image. The proposed architecture uses two consecutive convolutional layers to identify the features of the facial expressions and one fully connected layer to recognize the facial expressions. The proposed model achieves the accuracy which is 99.49% and 93.02% on CK+ and JAFFE datasets respectively. Which are the significant improvement over the recent works.