Comparative Analysis of Techniques for Recognising Facial Expressions

Abstract

Experts can predict the state of mind from human facial expressions and can be used in various human-machine interaction applications. In such applications the countenance recognition plays a crucial role. Understanding of Human be-haviour, predicting Mental health, and recognizing fabricated human expressions are some of the several applications of Automatic facial expression recognition system. Though predicting a human expression is an easy task for humans but it invites a competitive research work for computers to recognize expressions efficiently by extracting facial features from a digital image. As per the literature the basic approach used by the researchers for the automatic facial expression recognition system are based on appearance and geometry. Several techniques have been proposed by researchers for the Facial Expression Recognition. These techniques can be put into two broad categories, Con-ventional and Deep Learning methods. In this research work, the conventional methods Local Binary Pattern (LBP), Principal Component Analysis (PCA), Random Forest, and Support Vector Classifier (SVC) models are studied and implemented. From Deep learning category, a CNN model is developed and tested on the dataset FER2013. All the models implemented, tested on the same dataset are put into comparison with some existing techniques. From the analysis it can be concluded that Deep learning techniques perform better.