Computational model for emotion perception

Abstract

This study explores the interplay between gaze patterns and the saliency of facial features to improve emotion recognition. By analyzing where observers direct their gaze when identifying emotions and assessing the saliency of facial features that capture areas of image which grab attention, we aim to determine whether integrating these two factors can enhance predictive models of emotional perception. A dataset of facial stimuli portraying happy, angry, and neutral emotions was used, with regions of interest—eyes, nose, and mouth—carefully segmented for analysis. Saliency maps and gaze density maps were generated, and their correlations were computed to uncover patterns in emotional recognition. Results indicate that saliency highlights visually prominent areas but does not sufficiently differ entiate between emotions. While gaze patterns more effectively differentiate happy emotions, they remain insufficient for comprehensive emotion recognition. However, the combined correlation of saliency and gaze metrics demonstrates an improved capacity to identify emotional states, suggest ing the potential for integration in predictive modeling.