Combining convolutional neural networks for emotion recognition

Emotion is at the core of all human interaction, and thus as technology is becoming increasingly pervasive in daily life, emotion recognition is becoming increasingly relevant as well. It is an area that has a seemingly endless range of applications, from advertising and entertainment, to education and healthcare. Research has been done on detecting facial expressions from images using machine learning, and we expand on past work by applying new methods. In addition to training two convolutional neural networks individually, we also train new models that combine the two different neural networks at different stages of training. Our goal is to compare the results of early and late fusion of networks, and demonstrate that combining models leads to more accurate results for identifying emotion.