Each time through the wrinkles normal calculation, the calculated perturbation
is forced positive using fabs(), and at the end is always added to the original
normal. This will introduce a gradient, but it's not clear to me how it could be
removed without altering the overall character of the wrinkles normal.
