3-Dimensional facial expression recognition in human using multi-points warping

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Biometrics

Background:

• 3D facial expression recognition is challenging due to data acquisition issues and complex analysis.
• Human social communication relies heavily on facial expressions, easily identified by humans but difficult for machines.
• Existing methods struggle with 3D facial data due to lack of homology and intricate point digitization.

Purpose of the Study:

• To propose a novel method for 3D facial expression recognition using Multi-points Warping.
• To address challenges in 3D facial data acquisition and analysis for accurate machine identification.
• To develop a robust system for recognizing human facial expressions in 3D.

Main Methods:

• Utilized Multi-points Warping to create a reference template mesh for 3D facial landmarks.
• Applied the template mesh to target meshes from Stirling/ESRC and Bosphorus datasets.
• Employed Principal Component Analysis (PCA) for feature selection and Linear Discriminant Analysis (LDA) for classification, assessing localization error using Procrustes ANOVA.

Main Results:

• Achieved high recognition accuracy of 99.58% on Stirling/ESRC and 99.32% on Bosphorus datasets.
• Demonstrated superior performance compared to state-of-the-art methods in localization error validation.
• Visualized expression variations and identified 'Sad' as the expression with the lowest recognition accuracy.

Conclusions:

• The proposed Multi-points Warping method is robust for 3D facial expression recognition.
• The findings align with and improve upon current state-of-the-art results in the field.
• The method offers a significant advancement in computer vision for analyzing human facial expressions.