Performance improvement of Facial Expression Recognition Deep neural network models using Histogram Equalization and Contrast Stretching

Abstract

Human facial expression analysis has been in research for some time now, for its widespread application in healthcare, marketing, security, human-robot interactions, and many more. Deep neural networks are being used for developing facial expression recognition models. The quality and nature of the images used for model training are always deciding factors for the performance of the models developed. This paper describes how the application of two image enhancement methods, Histogram equalization, and Contrast stretching on the training images, improve the performance of neural network models used for facial expression. A basic convolutional neural network model and a transfer learning model are used in this study, for repeated performance analysis using multiple models. The experiments are carried out with three popular Facial Expression Analysis datasets, Real World Affective Faces, FER2013, and FER plus datasets to establish the results. It is found that applying histogram equalization or contrast stretching on the face images to be used in the model training, increases the accuracy of the model by an average of two percent.