Imagine that we have learned a point distribution for the shape of the human face. Now we see a new face where everything below the nose is occluded by a scarf. How could you exploit the model to...

Imagine that we have learned a point distribution for the shape of the human face. Now we see a new face where everything below the nose is occluded by a scarf. How could you exploit the model to estimate both the positions of the landmark points in the top half of the face and the landmark points in the (missing) bottom half of the face?