Estimation of Missing Facial Views

We describe a method for estimating new facial views from a single 2D face image. A model has been generated which incorporates two facial views, a frontal view and a 30 degree rotated view. Appearance models for each view of shape and texture have been combined using further principal components analysis to form a multiple view model. Presented with a novel face image in a frontal view, we can then use the multiple view model to generate an estimate of the same face as seen from a different viewpoint. The method is general and could also be applied to any object class in which the texture and shape of the constituent objects exhibits significant correlation.