MIFAD-Net: Multi-Layer Interactive Feature Fusion Network With Angular Distance Loss for Face Emotion Recognition.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Accurate human emotion recognition is crucial for advancing artificial intelligence.
• Subtle differences in facial expressions pose challenges for current computer-based emotion recognition systems.
• Existing models struggle with feature discrimination due to information loss during convolution and pooling.

Purpose of the Study:

• To propose a novel multi-layer interactive feature fusion network model with angular distance loss for improved facial emotion recognition.
• To enhance the model's ability to capture both global and local facial features at multiple scales.
• To address the challenges of subtle feature discrimination, information loss, and class separability in facial emotion recognition.

Main Methods:

• Designed a multi-layer and multi-scale module for extracting global and local facial emotion features.
• Implemented a hierarchical interactive feature fusion module with attention mechanisms to preserve feature information and enhance discriminative ability.
• Utilized an angular distance loss function to improve inter-class feature separation and intra-class feature clustering.

Main Results:

• The proposed MIFAD-Net achieved performance improvements of 1.02-4.53% over compared methods on the FER2013 dataset.
• The model demonstrated enhanced capabilities in discriminating subtle facial features and improving class separability.
• Ablation studies confirmed the effectiveness of the proposed modules and loss function.

Conclusions:

• The novel multi-layer interactive feature fusion network with angular distance loss significantly advances facial emotion recognition accuracy.
• The proposed model effectively addresses the limitations of existing methods in handling subtle expressions and feature information loss.
• This research contributes to the development of more sophisticated artificial intelligence systems capable of understanding human emotions.