Too Good to be True: Reevaluating the Role of Posed Stimuli in Emotion Recognition Tasks Using a High Frame Rate Database of Spontaneous and Posed Facial Expressions

Abstract

Emotion recognition studies predominately rely on posed stimuli, in which actors perform prototypical emotions on camera. These emotions are easy to recognize, but the stimuli are too good to be true representations of everyday emotional expressions. We hypothesized that spontaneous expressions of emotion seem more ambiguous than posed stimuli and are therefore harder to accurately identify.

To assess our prediction, we created a new video database of spontaneous and posed facial expressions of the six basic emotions. Spontaneous emotions were extracted from recorded informal conversations between the subject and experimenter. After the conversation, the subject was asked to perform the posed emotions. The footage was then coded into a database of 592 expressions, and hosted on our own platform: identifyemotions.com.

The platform was then used to conduct emotion identification tasks using subjects recruited from Amazon mTurk. We found that subjects were significantly better at identifying the basic emotions from posed stimuli than spontaneous stimuli. However, subjects were also generally more accurate at identifying the spontaneous expressions as real emotions than identifying the posed expressions as fake emotions.

Since people can perceive the difference between spontaneous and posed expressions, we must reevaluate previous emotion recognition studies that employed posed stimuli. Additionally, our database established a new precedent for standardization in facial expression databases in terms of video quality, frame rate, and expression variety.