Fit surface to large data set

I have a matrixe (size approx. 3000x3000) which includes z values of a surface. The values are very noisy and include NaN values (approx 10% of all values). What is the best way to find a fitted surface in matlab without the fitting toolbox?. The surface should be of the form ax^2 + by^2 + cx+ dy + exy + f.

Answer 1

[Meghana Dinesh]

Hey,

You can see Plane fit by Adrien Leygue. I have used this in my algorithms too for data-sets of approximately the same size.

Regards,

Meghana.

Answer 2

[ph0t0n]

thank you, Meghana, but a plane is not the best approximation for my surface. Does anyone know a method which allows quadratic terms ax^2 + by^2 + cx+ dy + exy + f as a surface model.