When recognizing complex shapes humans likely integrate information from the analysis of simpler components. Biological-motion recognition might also be based on an integration of simpler movement components. This hypothesis seems consistent with results from motor control showing that the control of multi-joint movements is based on simpler components, called synergies, which encompass a limited number of joints. We tested whether subgroups of joints as perceptual analogues of synergies define meaningful components for the perception of biological motion.

Extending an existing motion morphing technique (Giese and Poggio, 2000 International Journal of Computer Vision 38 59 - 73) we simulated point-light walkers with two different synergies (including the joints of the upper body and the lower body) by morphing between neutral walking and walks with different emotional styles (sad, angry, fearful). We separately varied the amount of information about the emotion conveyed by the two synergies. The percept of emotions was assessed by an expressiveness rating, and by a yes-no task requiring subjects to distinguish neutral and emotional walks (e.g. "neutral or sad?"). Subjects’ responses were fitted and predicted using Bayesian ideal-observer models that treat the contributions from the two synergies as independent information sources.

The morphed stimuli look very natural, even if only one synergy provides information about the emotion. As expected, ease of emotion recognition increases with the contribution of the emotional prototype to the morph. The contributions of the synergies to the overall perceptual judgement vary between emotions. Quantitative modelling shows that in most cases the emotion-recognition performance of the subjects can be predicted accurately by Bayesian ideal-observer models that integrate the emotion information provided by the two synergies in a statistically optimal way. We conclude that biological-motion recognition might be based on spatio-temporal components with limited complexity, and integrated in a statistically optimal fashion.