Machine Learning-Based Analysis of Facial Expressions and Emotion Regulation through Micro-Expressions

Abstract

Emotion Recognition through Facial Expression is an increasingly critical 
component of Human Computer Interface Technology (HCT), Mental Health 
Assessments and Remote Patient Monitoring. Traditional macro-expression 
based emotion recognition systems use facial expressions to assess a persons 
emotional status; yet, they do not capture the very subtle and involuntary 
movements of a persons face while experiencing emotion referred to as micro
expressions. Micro-expressions are generally short lived and may offer valuable 
insight to a person's actual emotional condition. This research proposes 
HybridMicroNet, a hybrid deep learning model to perform micro-expression 
detection that incorporates Resnet and Vgg-16 into one model in order to 
provide a better solution for the classification of images; it has been tested using 
data samples from many sources such as SAMM and CASME-II that contain 
micro-expression data. HybridMicroNet has great potential for deployment in 
numerous real world applications, such as Social Media Video Content and 
Clinical Interview Recordings, where there are often subtle or unlabeled 
emotional cues. For example, HybridMicroNet achieved 99.08% accuracy for 
the CASME-II dataset and 97.62% accuracy for the SAMM dataset. This 
demonstrates the ability of Machine Learning to recognize Emotions and Detect 
Mental Illness and highlights the growing interest in developing emotionally 
responsive systems.