Abstract:

We proposed a new method to synthesize facial expression. First, we represented face binary edge map with curve segment group. The differences of local texture beside the curve segments were measured, which are considered as the features of the images. Then we measured the deformation of curve segments and the change of the features when the samples in the face database had some kind of expression change, and learned the pattern behind it. Even when we only have the neutral expression sample,  following the pattern, we can still synthesize the curve segments and features of the sample for a certain type of expression change. Through this method, the sample space can be effectively extended, and a better performance is given by the recognition algorithm  when it faces the challenge of multiple expression change.

Key words: expression synthesis, edge extraction, curve fitting; polar moment of inertia