Humans have a tendency to express a standard and cardinal range of emotions through their facial expressions. There are seven basic raw emotions: happiness, anger, sadness, surprise, fear, neutral, contempt, and disgust. However, it is clear that most research focuses on the five primary emotions: happiness, surprise, anger, sadness, and neutrality. With the advancement of technology, it is now possible to automatically recognise a per- son’s emotions through videos and images, thanks to algorithms that detect, extract, and evaluate these facial expressions. The algorithm we propose and present is a hybrid face detection and feature extraction model that can detect the previously mentioned emotions in real time. The former is accomplished by means of the Haar cascades object detector, while the latter is accomplished by means of Deep Convolutional Neural Networks (DCNN). The hybrid method aids in the rapid initial face detection step, followed by the extraction of features (starting with generalised [low frequency] features and subsequently more specific [high frequency] features). On the FER 2013 dataset, the proposed model achieved an accountable training accuracy of 84.40% and a validation accuracy of 74.74%.

Convolutional Neural Networks, Facial Expres- sions, Real-Time Detection, Feature Extraction