Micro-Expression Recognition Using Color Spaces.

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Affective Computing

Background:

• Micro-expressions are involuntary facial cues vital for emotion detection and lie detection.
• Previous research highlighted the utility of CIELab and CIELuv color spaces for micro-expression recognition.
• Existing methods often rely on RGB or grayscale, potentially missing crucial color information.

Purpose of the Study:

• To propose and evaluate a novel color space model, tensor independent color space (TICS), for micro-expression recognition.
• To demonstrate the effectiveness of CIELab and CIELuv in micro-expression analysis.
• To explain the superior performance of these color spaces over RGB and grayscale.

Main Methods:

• Micro-expression video clips were modeled as fourth-order tensors, incorporating spatial, temporal, and color information.
• The RGB color dimension was transformed into the proposed TICS model, aiming for maximal color component independence.
• Dynamic texture histograms were computed for defined regions of interest (ROIs) based on the Facial Action Coding System.

Main Results:

• The TICS color space, combined with dynamic texture analysis, achieved higher accuracy in micro-expression recognition than RGB.
• CIELab and CIELuv also demonstrated improved performance in micro-expression recognition tasks.
• Experimental results on the CASME and CASME 2 databases confirmed the superiority of TICS, CIELab, and CIELuv over RGB and grayscale representations.

Conclusions:

• The proposed TICS color space offers a significant advancement for micro-expression recognition by maximizing color independence.
• CIELab and CIELuv are valuable color spaces for analyzing micro-expressions, outperforming traditional methods.
• Integrating advanced color spaces with dynamic texture analysis provides a more robust approach to understanding subtle facial expressions.