Interrelating physical feature of facial expression and its impression

The purpose of this work is to analyze human perception on the basis of a model of the interrelationship between physical feature of face and its emotional impression by using a unique neural network that is able to find mapping of non-linear relationship. Human's impressions of faces can be visualized on to the two or three dimensional semantic space that is nonlinearly mapped from facial physical parameters. Additionally, by obtaining the inverse nonlinear mapping, facial expressions can be reproduced from the semantic parameters. This approach does not only show good performance compared to the conventional statistical method, but also able to map a new data that are not used at the training phase to examine the appropriateness of the obtained mapping function.