Evaluation of Synthetic Faces: Human Recognition of Emotional Facial Displays

A first evaluation study of a database of kinetic facial expressions (DaFEx)

“…In this case, participants were asked to watch the DaFEx videos and to classify each one of them. For the data analysis, we used loglinear analysis [1] mainly following [8] and [9].…”

“…In the first place, Ekman's model has a high theoretical importance, coming with a rich set of constructs that are of immediate relevance for ECAs developers. Secondly, among the categorical models, it is the most used by researchers and developers; hence, it seemed appropriate to maintain the compatibility with most of the studies on the way facial expressions are perceived, both when produced only by humans [4], [6], [11], [17] and when comparing those of humans with those of synthetic agents [2], [8], [9], [16]. Finally, the alternative of resorting to dimensional approaches, as in, e.g., [19] and [20], did not seem practical for the purpose of setting up a benchmark for ECAs assessment: at least with respect to recognisability it is much easier to use discrete categories than continuous dimensions.…”

“…Other works -e.g., [2], [8], [9], [16] -have addressed the quality of facial displays, and those 'low levels' dimensions that arguably determine it: lip-speech synchronisation, facial gestures signalling emotions, or emphasis, punctuation and other discourse-level regulatory characteristics. However, despite the current interest in the topic, and the many efforts towards endowing ECAs with the capabilities of expressing emotional states, neither accepted benchmarks nor agreed methodologies are available to assess those dimensions.…”

Data-Driven Tools for Designing Talking Heads Exploiting Emotional Attitudes

“…Results from a preliminar experiment for the evaluation of the adequacy of facial displays in the expression of some basic emotional states, based on a recognition task, are presented in a paper in this volume, where also the potentials of the used evaluation methodology are discussed [19].…”

Preliminary Cross-Cultural Evaluation of Expressiveness in Synthetic Faces

“…They propose to measure the expressiveness of a synthetic face through the accuracy rate of human observers who recognise the facial expression, and to compare the expressions of the synthetic face with those of the "original" human face, upon which they are based. Similarities and dissimilarities between their methodology and ours are discussed in details in [2]. Section 2 of the paper illustrates how the test data were acquired and exchanged across sites.…”