Facial expression recognition in perceptual color space

Area of Science:

• Computer Vision
• Machine Learning
• Image Processing

Background:

• Facial expression recognition (FER) is crucial for human-computer interaction.
• Traditional FER methods often rely on grayscale images, potentially missing vital color information.
• Illumination variations and low-resolution images pose challenges to robust FER.

Purpose of the Study:

• To introduce a novel Tensor Perceptual Color Framework (TPCF) for enhanced facial expression recognition.
• To investigate the contribution of color information to FER performance.
• To evaluate the efficacy of different color spaces and feature extraction methods for FER.

Main Methods:

• Utilizing multi-linear algebra and tensor concepts to unfold color image components into 2-D tensors.
• Extracting features using Log-Gabor filters within various color spaces (RGB, YCbCr, CIELab, CIELuv).
• Employing the Mutual Information Quotient (MIQ) for feature selection and multi-class Linear Discriminant Analysis (LDA) for classification.

Main Results:

• Color information provides significant complementary data for improving FER robustness, particularly with texture characteristics.
• Perceptual color spaces (CIELab, CIELuv) outperform other color spaces for FER, especially under varying illumination.
• The TPCF framework demonstrates improved performance on low-resolution and illumination-variant facial expression datasets.

Conclusions:

• Color information is a vital component for advancing facial expression recognition systems.
• The TPCF offers a robust approach for leveraging color data in FER.
• Perceptual color spaces are recommended for developing more effective and resilient FER applications.