Automated measurement of spontaneous surprise

Humans exhibit a rich variety of facial expressions reflecting their emotional state. As pointed out by [6], automatically measuring facial expressions to infer emotional states is a challenging task, amongst others because there are no uniquely defined representational units of facial expressions for the basic emotions. In natural circumstances, spontaneous facial expressions tend to be subtle and highly context-dependent, which makes their formal representation very difficult. We have adopted an approach that performs computational analyses on video data acquired from carefully designed experiments for the measurement of facial expressions (see, e.g. [4]). This paper reports on our approach to measure spontaneous surprise. Two main obstacles in automatically measuring spontaneous surprise are of a behavioral and a technical nature. The behavioral obstacle is to elicit and identify spontaneous surprise. The technical obstacle is to develop a computational method to measure the facial expressions associated with spontaneous surprise. In order to deal with the first obstacle two of the authors (MS and JK) developed a novel experimental paradigm to elicit spontaneous surprise in humans. The paradigm is described in Section 2. The second obstacle is addressed by the Spontaneous Surprise Measurement method presented in Section 3. The setup for the experimental evaluation of the method is described in Section 4 and the results are reported in Section 5. Finally, Section 6 is a concluding discussion.