Parameter optimization of histogram-based local descriptors for facial expression recognition

Area of Science:

• Computer Vision
• Machine Learning
• Pattern Recognition

Background:

• Facial expression recognition (FER) requires robust feature descriptors.
• Descriptors must handle variations in scale, illumination, pose, and noise.

Purpose of the Study:

• To evaluate the effectiveness of spatially modified local descriptors for FER.
• To investigate the impact of face registration on recognition accuracy.
• To optimize parameters for four key local descriptors: HOG, LBP, CLBP, and WLD.

Main Methods:

• Comparative analysis of feature extraction from registered versus non-registered faces.
• Systematic optimization of parameters for Histogram of Oriented Gradients (HOG), Local Binary Patterns (LBP), Compound Local Binary Patterns (CLBP), and Weber's Local Descriptor (WLD).

Main Results:

• Face registration significantly improves the recognition rate in FER systems.
• Optimized parameter selection enhances the performance of local descriptors.
• The proposed methods show competitive results compared to state-of-the-art approaches.

Conclusions:

• Face registration is a crucial preprocessing step for accurate FER.
• Careful parameter tuning is essential for maximizing the potential of local feature descriptors.
• Spatially modified local descriptors, when optimized, offer robust feature extraction for facial expression recognition.