A Convolutional Neural Network for Emotion Assessment and Recognition

Abstract

With the vast and rapid expansion of the digital age, artificial intelligence is becoming a key factor in the technological realm. As humans become more immersed into technology, computer scientists and researchers are innovating increasingly complex machine and deep learning algorithms to match consumer need. These programs have a broad range of applications, not limited to computer vision, fraud detection, auto-correction, and more. An additional application - and the focus of this research - is emotion recognition and assess-ment. The concept of creating an emotion recognition system is one that has been researched heavily since the late 1990s. Monitoring emotional disorders in everyday environments such as a classroom or a hospital could be done in an almost fully automated manner with the correct improvements on emotion recognition algorithms. They can also be used to help assess and provide solutions for rectifying or managing unwanted emotional behavior. An emotion assessment system can be realistically built by following well-documented machine and deep learning principles, and error analysis as well as precision readings can be performed with acceptable accuracy. The focal point of this research is to create a Deep Convolutional Neural Network (DCNN) model that categorizes 7 different human facial emotions. The model is trained, tested, and validated using the preprocessed FER-2013 dataset.