Learning a single active face shape model across views

An algorithm is described for modelling and recovering the shape of a face varying from the left to the right profile views. The method is based on a multi-view nonlinear model that utilises 2D view-dependent context without explicit reference to 3D structures. The model can cope with large nonlinear shape variations and inconsistent facial feature landmarks between wide varying views. For nonlinear model transformation, we adopt Kernel PCA based on the concept of support vector machines.