A Perception CNN for Facial Expression Recognition.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Convolutional Neural Networks (CNNs) are widely used for facial expression recognition (FER).
• Existing CNNs may overlook the importance of facial segmentation in accurately recognizing expressions.
• There is a need for models that can capture both local facial cues and global facial structure for improved FER.

Purpose of the Study:

• To propose a novel Perception Convolutional Neural Network (PCNN) for enhanced facial expression recognition.
• To improve the sensitivity of FER systems to subtle facial changes.
• To achieve superior performance on various FER benchmarks by integrating local and global facial information.

Main Methods:

• The proposed PCNN utilizes five parallel networks to learn local facial features from eyes, cheeks, and mouth.
• A multi-domain interaction mechanism fuses local sensory organ features with global facial structural features.
• A two-phase loss function is designed to ensure the accuracy of extracted information and reconstructed facial images.

Main Results:

• The PCNN demonstrates superior performance on multiple FER benchmarks, including CK+, JAFFE, FER2013, FERPlus, RAF-DB, and the Occlusion and Pose Variant Dataset.
• Experimental results validate the effectiveness of the proposed PCNN in capturing subtle facial variations.
• The integration of local and global features significantly enhances FER accuracy.

Conclusions:

• The developed PCNN offers a significant advancement in facial expression recognition.
• The PCNN's ability to process both local and global facial information leads to robust and accurate expression classification.
• The proposed model provides a promising direction for future research in FER systems.