Micro-Expression Recognition Based on Pixel Residual Sum and Cropped Gaussian Pyramid.

Area of Science:

• Computer Science
• Artificial Intelligence
• Biomedical Engineering

Background:

• Facial micro-expression (ME) recognition is crucial for understanding human emotions but is challenging due to detection difficulties and environmental interference.
• Existing methods struggle to accurately capture subtle facial movements essential for reliable ME recognition.

Purpose of the Study:

• To develop novel preprocessing methods to enhance the detection and recognition of facial micro-expressions.
• To improve the robustness of ME recognition systems against environmental disturbances.

Main Methods:

• Proposed two preprocessing methods based on Pixel Residual Sum for unit pixel displacement and environmental resistance.
• Introduced a Cropped Gaussian Pyramid with Overlapping (CGPO) module for multi-resolution image processing.
• Utilized a depthwise convolution-based CNN with progressively increasing channels for feature extraction and fusion with position embedding.

Main Results:

• The proposed Pixel Residual Sum preprocessing methods effectively resist environmental interference and facilitate subtle facial feature extraction.
• The CGPO module and CNN architecture successfully extracted and fused preliminary features for improved recognition.
• Experimental results demonstrate superior performance of the proposed methods and model compared to existing well-known methods.

Conclusions:

• The novel preprocessing techniques and CGPO module significantly advance facial micro-expression recognition capabilities.
• The developed model offers a more robust and accurate solution for detecting genuine human emotions through subtle facial cues.