Abstract: We present underpinning neural structures to represent various components of emotional dispositions. After a description of simple models of the dynamics, an even simpler multi-layer perceptron model is presented with three outputs. This leads to a high level of success in recognition of emotional dispositions from a database of faces, as well as an interpretation in terms of the underlying neural system. By extending the discrete classification approach to continuous variables in a three-dimensional state-space the recognition performance can be improved and the causes for classification errors can be studied.

Key-Words: facial analysis, emotion recognition
IMACS/IEEE CSCC'99 Proceedings, Pages:5341-5346