Robust Human Face Emotion Classification Using Triplet-Loss-Based Deep CNN Features and SVM.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Human facial emotion detection is complex due to high inter-class variance and diverse emotional expressions.
• Accurate facial emotion classification remains a challenge for existing machine learning models.

Purpose of the Study:

• To propose a novel and intelligent approach for accurate human facial emotion classification.
• To enhance the performance of facial emotion recognition systems.

Main Methods:

• A customized ResNet18 model integrated with a triplet loss function (TLF) was developed.
• Transfer learning was employed, followed by Support Vector Machine (SVM) classification.
• RetinaFace was utilized for face detection and extraction, with ResNet18 trained on cropped faces using triplet loss.

Main Results:

• The proposed method achieved superior performance compared to state-of-the-art (SoTA) methods on the JAFFE and MMI datasets.
• Accuracies of 98.44% and 99.02% were recorded on the JAFFE and MMI datasets, respectively, for seven emotions.
• Further fine-tuning is required for the FER2013 and AFFECTNET datasets.

Conclusions:

• The integration of triplet loss with ResNet18 and SVM offers a robust pipeline for facial emotion classification.
• The developed method demonstrates significant potential for advancing the field of computer vision-based emotion recognition.
• Future work will focus on optimizing the approach for more challenging datasets like FER2013 and AFFECTNET.