Facial Feature Extraction Using a Symmetric Inline Matrix-LBP Variant for Emotion Recognition.

Area of Science:

• Computer Vision
• Machine Learning
• Pattern Recognition

Background:

• Local Binary Patterns (LBP) are crucial visual descriptors in computer vision.
• Numerous LBP variants exist, highlighting the need for improved feature extraction methods.
• Efficient facial expression recognition (FER) requires robust visual descriptors adaptable to varying conditions.

Purpose of the Study:

• Introduce a novel visual descriptor, Symmetric Inline Matrix-Local Binary Patterns (SIM-LBP).
• Evaluate SIM-LBP's efficacy in extracting facial features for FER.
• Compare SIM-LBP performance against traditional LBP and its variants.

Main Methods:

• Developed the SIM-LBP descriptor using a novel matrix generation technique.
• Applied SIM-LBP transformation to images in the JAFFE dataset.
• Trained a Convolutional Neural Network (CNN) model using SIM-LBP transformed images for FER.

Main Results:

• SIM-LBP demonstrated superior performance in extracting facial features, particularly under diverse lighting conditions.
• The CNN model trained with SIM-LBP achieved higher recognition accuracy, precision, recall, and F1-score compared to baseline LBP methods.
• Comparative analysis confirmed SIM-LBP's effectiveness against state-of-the-art methods.

Conclusions:

• SIM-LBP is a highly effective visual descriptor for facial expression recognition.
• The proposed SIM-LBP method offers significant improvements over existing LBP variants.
• SIM-LBP has potential applications in identifying mental states and predicting mood variations from facial images.