Different views of facial expressions: an image sequence dataset

Abstract

Here we describe a face expression dataset containing image sequences showing different views of facial expressions as they develop from neutral to full onset. We recruited actors (11 female, 8 male) from our Drama Department and elicited 5 facial expressions (anger, disgust, fear, happiness and sadness) in a controlled lighting environment. We used a multi-camera array to gather high definition (1920×1080 pixels) uncompressed RGB image sequences (2 seconds at 25 Hz) simultaneously from 5 viewpoints (−45, −22.5, 0, 22.5 and 45 degrees), capturing the onset and development of the facial expressions. Actors faced the 0 degree camera. The angle of rotation was parallel to the ground plane and the cameras were equidistant from the actors' faces. In order to aid colour-adjustment of the images for (for example) daylight, we also gathered pictures of the actors using a calibrated digital camera and measured the colour characteristics of our scene using a spectral photometer (sampled from 380 to 760 nm in 5 nm steps). We show how our dataset can be combined with morphing techniques to produce finely graded changes in emotional expressions suitable for use in psychophysical experiments. Such sequences closely follow the actual development of the expressions and avoid morph-induced artifacts that may occur when one morphs between a neutral face and a full-blown expression. When combined with standard non-parametric staircase techniques this offers a powerful tool to assess sensitivity to facial expressions. In addition, we present data using facial emotion ratings and identification to provide a direct comparison of the facial expressions of our 19 actors with those of 10 actors from the Ekman and Friesen (1976) image set. To our knowledge our dataset provides a unique resource that can potentially be used by a wide variety of researchers interested in the perception of facial expression.

Footnotes

 This research was supported by the British Academy