Robust facial expression recognition system in higher poses

Area of Science:

• Computer Vision
• Biometrics
• Human-Computer Interaction

Background:

• Facial expression recognition (FER) is crucial for applications in security, neuroscience, and psychology.
• Current FER systems struggle with accuracy in severe head poses, limiting their real-world applicability.
• Addressing this challenge is vital for advancing FER technology.

Purpose of the Study:

• To improve the prediction accuracy of facial expression recognition (FER) specifically in cases of severe head poses.
• To introduce a robust 3D head-tracking algorithm integrating an ellipsoidal model, ensemble AdaBoost, and saturated vector machine (SVM).

Main Methods:

• Utilized an ellipsoidal tracking model for robust feature tracking across video frames.
• Employed Gabor filters for the extraction of visible expressive facial key points.
• Implemented an advanced ensemble algorithm (Ada-AdaSVM) for effective feature selection and classification.

Main Results:

• The proposed technique demonstrated outstanding overall performance across five diverse facial expression databases (Bosphorus, BU-3DFE, MMI, CK+, BP4D-Spontaneous).
• Significant improvements in recognition accuracy were observed, particularly under challenging severe head pose conditions.

Conclusions:

• The developed 3D head-tracking algorithm effectively overcomes the limitations of severe head poses in FER.
• This robust approach offers a promising solution for enhancing the reliability and accuracy of facial expression recognition systems.