Automated Facial Expression Recognition

This booklet introduces four methodological tutorials on how to implement neural networks to predict human emotions using detailed Python codes and a walk-through example. The booklet uses the emerging field of automated facial expression recognition (FER). The first two tutorials show step-by-step instructions, supported by theories, on how to implement neural network from scratch for image classification using multinomial logistic regression and shallow artificial neural network (ANN). The last tutorials rely on third-party neural network libraries to perform classification tasks using deep neural network (Keras Deep Learning library) and convolutional neural networks (CNN). The objective is to simplify FER to users.

Our validation metrics shows that the accuracy of FER neural networks models are inevitably tied the classifiers and the developer hyper-parameter choices, computer power, and the quality, size, and diversity in data to optimize machine learning and training and eliminate racial and gender bias that might lead to false positives.