Enhancing Mouth-Based Emotion Recognition Using Transfer Learning.

Area of Science:

• Computer Science
• Artificial Intelligence
• Biomedical Engineering

Background:

• Limited datasets and simulated expressions hinder accurate mouth-based emotion recognition.
• Transfer learning offers an efficient solution by leveraging existing models for new tasks.
• Real-world emotion recognition requires dynamic adaptation to various scenarios.

Purpose of the Study:

• To investigate the efficacy of transfer learning for mouth-based emotion recognition.
• To develop a dynamic emotion recognition system adaptable to new and modified situations.
• To evaluate the performance of mouth-based emotion recognition against full-face recognition.

Main Methods:

• Utilized transfer learning to fine-tune convolutional neural networks (CNNs) with limited emotional data.
• Employed an extensive, literature-recognized dataset for face-based emotion recognition for testing.
• Compared the accuracy of the proposed mouth-based approach with traditional full-face recognition methods.

Main Results:

• Transfer learning significantly improved CNN performance accuracy in mouth-based emotion recognition.
• The proposed method demonstrated consistent performance, even when the full face was not optimally visible.
• Accuracy loss compared to full-face recognition was minimal and largely compensated by robust performance.

Conclusions:

• Mouth-based emotion recognition using transfer learning is highly effective, especially with limited data.
• The approach offers dynamic adaptability, crucial for real-world healthcare and assistive technology applications.
• Mouth detection is a vital component in the complex process of accurate emotion recognition.