Performance Evaluation of Deep Learning Models for Face Expression Recognition

Abstract

Facial expression recognition presents a significant challenge in computer vision, crucial for various applications like human-computer interaction and emotion analysis. Despite its importance, accurately discerning emotions from facial images remains complex due to factors such as lighting variations, pose differences, and subtle expression nuances. In this study, we aim to comprehensively evaluate five deep learning models - CNN, VGG16, Inception V3, MobileNet V2, and DenseNet121 - utilizing the CK+ dataset. Our research seeks to clarify the objectives and contributions early, emphasizing the significance of facial expression recognition. We provide an overview of the paper’s structure to guide the reader through the logical progression of ideas. The background and related work section reviews existing literature, highlighting recent advancements and identifying research gaps. The methodology details dataset characteristics, preprocessing steps, and model architectures, followed by the experimental results section presenting performance metrics and comparisons. The discussion interprets results, analyzing model strengths and weaknesses while considering practical implications and future research directions. Finally, the conclusion summarizes key findings and emphasizes the study’s significance, suggesting avenues for further exploration. Throughout the paper, clarity, readability, and grammatical accuracy are maintained, supported by visual aids like tables or diagrams where necessary to enhance comprehension.