Emotion recognition from static images

Abstract

Nowadays, automation using artificial intelligence is being introduced throughout the whole spectrum; From households to large enterprises. This report describes an emotion recognition model with the main aim to use artificial intelligence to classify emotions from static facial images. This will allow automatic machine classification of fundamental facial expressions. Multiple research endeavours have been concluded in effort of obtain good performance metrics of emotion recognition in un-posed environments. Despite these works, it still remains an open research problem as it has not reached the quality of other classifiers. The challenge lies within the amount of data needed to train such deep learning architectures and the respective quality. In this work, different Convolutional Neural Network (CNN) architectures will be trained using transfer learning methodologies, for the purpose of recognizing emotion from static images. Furthermore, two different training sets with distinct properties will be utilized to observe any discrepancies between the two. The main aim is to discover in more detail the primary challenges of this research problem, and implement a deep learning architecture for classification. The implemented, VGG-Face and a modifiedResNet-18architecture achieved a top-1 accuracy of 71.2% and 67.2% on FER2013 Dataset respectively. While results on the Affect Net Data set were of 58.75% with VGG-Face and 55.2% with ResNet-18 modified. This demonstrated that transfer learning from a close-point, is a very effective method to obtain better performance without the need to fully train a network. The discrepancies between the two results are analysed through confusion matrices.