Bartlett, Marian Stewart
Coauthors(s): Marian Stewart Bartlett and Javier R. Movellan Institute for Neural Computation, UCSD Department of Cognitive Science, UCSD
U.C. San Diego
Institute for Neural Computation
U.C. San Diego Institute for Neural Computation 9500 Gilman Drive, Dept. 0523 La Jolla, CA 92093-0523
www.cnl.salk.edu/~marni

Lying Faces

Spontaneous facial expressions differ from posed expressions of emotion. Genuine smiles, for example, can be differentiated from posed smiles by the contraction of orbicularis oculi, a muscle that circles the eye (Ekman, Friesen, & O'Sullivan, 1988). In genuine smiles, the orbicularis oculi is contracted, raising the level of the cheek and producing crows-feet wrinkles at the outer corners of the eyes, whereas in false smiles, this action is frequently omitted. In addition, there are differences in the facial symmetry of posed versus false expressions, and differences in their dynamics. Genuine expressions have a faster and smoother onset and offset than posed expressions, and feature apex coordination in which the activations of all of the facial muscles involved reach their peaks simultaneously. Methods for automatically discriminating spontaneous from posed facial expressions in image sequences are presented and tested on the task of discriminating the two classes of smile. We compare the results obtained with two different image processing approaches (Gabor energy filters, and ICA energy filters) and 3 different recognition engines (multilayer perceptrons, Gaussian classifiers, and hidden Markov models).