Emotion Recognition Using Deep Learning

Recognition of human emotions such as anger, disgust, fear, happiness, sadness, surprise and neutrality through facial expression is one of the important research topics in human computer interaction. There are several methods in the literature to identify emotions using machine learning and Artificial Intelligence techniques. This paper investigates the suitability of deep learning methods for the classification of emotions. Two classes of deep learning models: one belonging to non-pretrained models such as ConvNet, LeNet and the other belonging to pretrained models such as VGG 19 and MobileNet have been experimented for the extraction of features followed by Softmax for the classification. The classification accuracy of 95% is obtained with LeNet whereas MobileNeT gives the best accuracy of 96% on the database.