Convolutional Neural Network Architecture for Facial Emotion Recognition on Raw FER2013 Dataset

Abstract

Deep Learning now a days, is being used extensively in all the Compute Vision related researches and applications.  A Convolutional Neural Network has been developed for facial emotion recognition task. In this study, the developed network is trained to classify every facial imagewhich are present in the dataset into one of the seven facial emotions. The Proposed Convolutional Neural Network architecture achieved an accuracy of 65.05% on the FER2013 dataset without any image augmentation. The model was developed using MATLAB R2018a and was run on NVIDIA GeForce 830M Graphic Processing Unit to expedite the training and classification process.