Emotion Recognition from Facial Expressions in Children and Adults Using Deep Neural Network

Abstract

This paper explores the emotion recognition techniques in children and adults and identifies the differentiating features. There is decent literature reported on facial emotion recognition (FER) in adults but limited focus on children. We propose the use of a light-weight model incorporating a geometric-based approach known as facial landmarks combined with a simple Deep Neural Network (DNN) architecture. The findings from this research led to a detailed analysis to determine the most expressive features on the human face, the features sufficient or necessary for FER, and how this set of necessary features varies in children and adults. Furthermore, this paper aims to establish an estimate of the optimal number of facial landmark points that are sufficient to obtain appreciable results for emotion classification. Child Affective Facial Expression (CAFE) dataset and the Extended Cohn–Kanade (CK+) dataset have been used in this work.