Deciphering Facial Expressions and Emotion Regulation using Micro expressions and Machine Learning

Abstract

Facial expressions-based emotion recognition has emerged as an essential domain across human-computer interaction, affective computing, mental health detection, and healthcare monitoring systems. While traditional macro-expression analysis captures overt emotional states, it often overlooks the subtle, involuntary, and rapid facial changes known as micro-expressions. These subtle variations provide deeper insights into an individual s genuine emotional state. This thesis presents a comprehensive framework architecture to recognize and classify emotions through facial micro-expressions and deep learning algorithms along with Action Units (AUs)-based enhancement strategies. The framework is particularly geared toward applications in real-world, uncontrolled environments such as social media videos and clinical interviews contexts where emotional cues are subtle, inconsistent, and labels are often unavailable. newlineThe proposed work is structured around two primary models: HybridMicroNet, a fusion of ResNet and VGG-16 architectures optimized for micro-expression feature extraction, and an Enhanced AUs-based model. It integrates domain knowledge of Facial Action Coding Systems to map AUs to newly defined categories of novel micro-expressions. These models were trained and evaluated on benchmark datasets, including SAMM and CASME-II, and subsequently deployed on unlabeled real-world datasets sourced from YouTube. The proposed framework performance evaluated in terms of precision, recall, F1-score, accuracy, and confusion matrix. newline

Description

Keywords

Citation

item.page.endorsement

item.page.review

item.page.supplemented

item.page.referenced