Facial Emotional Expression Recognition Using Hybrid Deep Learning Algorithm

Abstract

Facial expression is the most common way to demonstrate individual emotional state. People intend to understand other people’s emotional state by observing their interactive partner’s facial expression. However, there are some limitations using the mentioned approach regarding the individual observative capability and the interactive partner’s privacy. Hence, the facial emotional expression recognition system based on Convolutional Neural Network (CNN) was employed. The most reliable approach is utilizing the state-of-the-art, such as Inception Net, ResNet, and VGG which had been developed to excel in their specific feature extraction approach. In addition to the mentioned models, there is also a common usage model, such as Convolutional Auto Encode (CAE) which is capable of high-efficient noise reduction. Then, some more advanced models were developed based on the concept of the initially state-of-the-art models, such as U-Net to perform high-performance image segmentation using feature fusion and transposed convolution technique. In this paper, the hybrid deep learning algorithm based on CNN and CAE is developed using the significant features from the mentioned state-of the-arts and 2 combinations of the modified CNN model to predict human emotional state. The experimental result shows the proposed model which achieves the predictive accuracy of 88%