The derivative of component j of the output point, with respect to x_d
(the dth component of the vector) is:
\sum_i D[ d, l ] G'( r ) ( x_j - l_ij ) / ( sqrt( N_l(p) ))
where: D is the D matrix i indexes the landmarks N_l(p) is the squared
euclidean norm between landmark l and point p l_ij is the jth component
of landmark i G' is the derivative of the kernel function.
Copies data from the W matrix to the D, A, and b matrices which represent
the deformable, affine and translational portions of the transformation,
respectively.