An Across-Target Study on Visual Attentions in Facial Expression Recognition

Area of Science:

• Cognitive Science
• Computer Vision
• Human-Computer Interaction

Background:

• Automatic expression recognition research relies on advanced devices like eye-trackers for observing human expression processing.
• Current studies face limitations in across-target analysis due to discrepancies in target faces, hindering the discovery of latent eye-behavior patterns.

Purpose of the Study:

• To explore attention patterns in expression recognition by enabling across-target analysis.
• To overcome limitations of previous studies by utilizing correspondences between different facial targets.

Main Methods:

• Developed a method to map eye-tracking fixations from different targets onto a synthetic face, creating an across-target fixation map.
• Tokenized fixation maps using areas of interest (AOI), analyzed in receiver operating characteristic (ROC) space, and modeled using linear regression.
• Compared attention patterns using Pearson's correlation.

Main Results:

• Achieved an averaged correlation values ranging from 0.60 to 0.86.
• Demonstrated significant similarity in eye-movement patterns between subjects when recognizing identical expression classes.

Conclusions:

• The proposed across-target analysis method effectively reveals consistent attention patterns in expression recognition.
• Findings suggest that underlying eye-behavior patterns are shared across subjects for specific facial expressions, advancing the field of automatic expression recognition.