To be robust to skin color and luminance change, our wrinkle
synthesis runs in the gradient domain. Basically, there are three
types of images used for both the input and the example face: the
RGB image I, the gradient image G and the height image H, where
the gradient image G stores ∂I/∂x and ∂I/∂y in its two channels
and the height values are stored in H. Given an input image If , we
first convert it into a gradient image Gf . Afterwards, we synthesize
a gradient image Gs using the patches in the example gradient
image Ge and compute a height image Hs using the correspondence
between the example gradient image Ge and height image He. Hs
is then used to displace the vertices on the face model to form
wrinkles. The algorithm pipeline is shown in Fig. 2. In the following,
we first describe how to generate the exemplars and then detail the
wrinkle synthesis algorithm.