Classifying Emotions using Convolutional Neural Networks

Deep Learning achieves surprising performance in many real-world tasks. In the past few years, image-based classification has been extensively studied and implemented. The current technology is very useful for achieving simple recognition of human expressions and classification. However, since there may be great individual differences between expressions evoked by different human beings, which may interfere with the extraction of image features, how to improve the accuracy of recognition and classification is still a very important issue. In this paper, a CNN model is trained on a dataset, Fer2013, of 28,709 photos with 7 human tagged emotions, by improving VGG-16 using Keras. The training results of this model are discussed and analyzed.