Authors :

Tan Wei Ling and Mohd Amirul Arif

Address :

Reseach Assistant, Department of Computer Science, Universiti Malaya, Malaysia

Professor, Department of Artificial Intelligence, Universiti Teknologi Malaysia, Malaysia

Abstract :

AIn recent years, with the increasing demand for surveillance and security systems in public places and for understanding human behaviour, an automated expression understanding from video is receiving significant attention. Video-based facial expression analysis has significantly increased and is applied to different states of individuals in healthcare, security, and human-computer interaction. The potential of accurately identifying emotions from facial expressions in video motion can enhance communication and user experience. However, extracting facial expressions in dynamic motion, like raising an eyebrow, is a complex mechanism compared to actions in static positions. Thus, with the development of deep learning models, the research introduces a Global-Local feature-assisted Dynamic Convolutional Neural Network (Glo-DynaCNN) for understanding facial expression changes across dynamic video frames. This network model's main objective is to dynamically extract global features and temporal characteristics to analyze the expressions in different frames. At the same time, local features focus on critical facial regions, including mouth and eye position, which extract expression variations. The proposed model dynamically improves facial expression identification accuracy with these feature extractions. For this, the research employed publicly available video facial expression datasets that demonstrate, with annotations for happiness, sadness, anger, surprise, fear, and disgust expressions, achieving significant improvements over existing recognition models. The model's performance is evaluated using metrics like identification accuracy, precision, recall, F1-score, confusion matrix, and Mean Square Error (MSE) with an improved analysis rate

Keywords :

Global Feature Analysis, Local Feature Analysis, Facial Expression Identification, Feature Extraction, Video Analysis.