Purchase this article with an account.
Martin Giese, Lars Omlor, Claire Roether; Learning and perceiving informative spatio-temporal components from emotional body expressions. Journal of Vision 2006;6(6):795. https://doi.org/10.1167/6.6.795.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Humans are able to communicate emotions through their posture, and through the dynamics of their body movements. Features that convey critical information about specific emotions have only rarely been studied, typically based on perceptual ratings (e.g. Montpare et al., 1987; Meijer, 1991; Wallbott, 1998). A more accurate characterization of informative features can be obtained by a statistical analysis of trajectories of emotional body movements. Studies on image statistics have shown that Independent Components Analysis (ICA) allows to extract highly informative features for the perception of natural images and faces. This motivates the question whether related approaches allow to extract informative spatio-temporal components for the visual perception of emotional body expressions.
METHOD: Our analysis was based on motion capture data from actors performing actions with different emotional affects. This data was analyzed with existing ICA methods, and applying a new algorithm that combines non-negative ICA with feature extraction by sparse regression. We extracted spatio-temporal components that contributed maximally to the approximations of trajectories with different emotional styles. In a psychophysical experiment, we tested the relevance of these components for the visual emotion categorization.
RESULTS: Opposed to standard ICA, the new algorithm extracts a small number of spatio-temporal components that are specific for individual emotions. These components seem to correlate with features that are important for the categorization of emotional body expressions. Informative features for visual recognition might thus reflect distinctive components in motor patterns that cannot be extracted with basic algorithms, like PCA or standard ICA.
This PDF is available to Subscribers Only