Realtime Facial Expression Recognition using Deep Learning

Abstract

Facial expression is the key to determine the physiological and psychological behaviour of any subject or crowd. Although it is easier for human beings to determine facial expressions, it's certainly a challenging task for machines. Automatic facial emotion recognition (FER) is an emerging research area in image classification. The availability of data as well as the ease of processing has opened up new possibilities of research with better accuracy. With the increase in computing capability deep learning processes have become a highly suitable technique in determining FER. As part of this paper, we have developed a deep convolutional neural network (CNN) which is used to classify the input images into seven different emotions as part of the study. We have used the FER2013 dataset images from the Kaggle website [1] to train our model and testing is done using still images as well as real-time surveillance data obtained from a webcam. A simple UI is used to display the test subject's real-time video along with the predicted labels during the time of video capture. The proposed neural networks are implemented using TensorFlow.