图像变形MLSD2DTransform---matlab自带的函数，其中内嵌的子函数PrecomputeAffine预先计算了权重，对三维数据不可！要改造

clear all
syms a1 a2 a3 a4 b1 b2 b3 b4 d1 d2 d3 d4 M1;
a=[a1,a2,a3,a4];
b=[b1,b2,b3,b4];
d=[d1,d2,d3,d4];
det = a.*d - b.^2//det会是一个向量，每个元都是a1d1-b1b1这样类似的元。

Ia = d./det
Ib = -b./det
Id = a./det

F1 = [sum(M1.[Ia;Ib],1);sum(M1.[Ib;Id],1)]

运行结果-------------------------------------------------------------
det =

[ - b1^2 + a1d1, - b2^2 + a2d2, - b3^2 + a3d3, - b4^2 + a4d4]

Ia =

[ d1/(- b1^2 + a1d1), d2/(- b2^2 + a2d2), d3/(- b3^2 + a3d3), d4/(- b4^2 + a4d4)]

Ib =

[ -b1/(- b1^2 + a1d1), -b2/(- b2^2 + a2d2), -b3/(- b3^2 + a3d3), -b4/(- b4^2 + a4d4)]

Id =

[ a1/(- b1^2 + a1d1), a2/(- b2^2 + a2d2), a3/(- b3^2 + a3d3), a4/(- b4^2 + a4d4)]

det =

[ - b1^2 + a1d1, - b2^2 + a2d2, - b3^2 + a3d3, - b4^2 + a4d4]

Ia =

[ d1/(- b1^2 + a1d1), d2/(- b2^2 + a2d2), d3/(- b3^2 + a3d3), d4/(- b4^2 + a4d4)]

Ib =

[ -b1/(- b1^2 + a1d1), -b2/(- b2^2 + a2d2), -b3/(- b3^2 + a3d3), -b4/(- b4^2 + a4d4)]

Id =

[ a1/(- b1^2 + a1d1), a2/(- b2^2 + a2d2), a3/(- b3^2 + a3d3), a4/(- b4^2 + a4d4)]

F1 =

[ (M1d1)/(- b1^2 + a1d1) - (M1b1)/(- b1^2 + a1d1), (M1d2)/(- b2^2 + a2d2) - (M1b2)/(- b2^2 + a2d2), (M1d3)/(- b3^2 + a3d3) - (M1b3)/(- b3^2 + a3d3), (M1d4)/(- b4^2 + a4d4) - (M1b4)/(- b4^2 + a4d4)]
[ (M1a1)/(- b1^2 + a1d1) - (M1b1)/(- b1^2 + a1d1), (M1a2)/(- b2^2 + a2d2) - (M1b2)/(- b2^2 + a2d2), (M1a3)/(- b3^2 + a3d3) - (M1b3)/(- b3^2 + a3d3), (M1a4)/(- b4^2 + a4d4) - (M1b4)/(- b4^2 + a4d4)]

从最后的F1的结果的前两列来看，它进行了取矩阵的逆矩阵的操作。
[ (M1d1)/(- b1^2 + a1d1) - (M1b1)/(- b1^2 + a1d1)
[ (M1a1)/(- b1^2 + a1d1) - (M1b1)/(- b1^2 + a1d1)
矩阵[a1,b1;b1,d1]的行列式作为了分母，
与M1实施了点乘法，即使M1为向量，只需要取第一个分量相乘。