Image size reveals perception biases of similarity among facial expressions of emotion

摘要

Recognizing facial expressions of emotion is important in social communication. Humans have the ability to recognize emotions from faces represented by a small number of pixels or at large distances. Where is the limit? Is this limit the same for all expressions of emotion? Or, are we more tuned to reading some specific emotions? We investigate these questions using six basic facial expressions of emotion (happiness, sadness, anger, fear, surprise, and disgust) in addition to neutral. Face images scaled in five different sizes encompassing 10x15 to 160x240 pixels (at increases of factors of 2) are presented in two emotion labeling tasks. Three important aspects of emotion recognition emerge from our study: 1) Recognition accuracy increases nonlinearly with image size in all expressions. 2) Happiness, surprise, disgust and neutral can be recognized at very small sizes (10x15 pixels), whereas fear, sadness and anger cannot (requiring images of at least 40x60 pixels). 3) At low resolutions, there is an asymmetric ambiguity in recognizing expressions of emotions a?? e.g., sadness is perceived as more similar to neutral than anger, while anger is most often confused with sadness; fear is more often misclassified as surprise than disgust, while disgust is typically misinterpreted as fear. This asymmetry is eliminated as the size of the image increases.