Get derivatives of a noisy surface

Hi everyone,

I still have trouble in getting derivatives of a surface. Is there any method to do this as the data are noisy? For the data z=f(x,y), how can I do the smoothing/ regularization to get the reliable derivatives. Someone suggested that regularizing the differentiation process to avoid the noise amplification of finite-difference methods. Is there any way to do that in matlab?

I found some ways to smooth and get derivatives for a curve fitting like z=f(x), but I don’t know how to deal with the 3D data.

Any suggestion? Your answer will be greatly appreciated.

Cheers Hui

[Walter Roberson]

This Question continues http://www.mathworks.com/matlabcentral/answers/8692-get-derivatives-from-noisy-data

[Zhenhui]

yes, but i still didnt find ways to solve it...T-T

[Zhenhui]

sure, thanks for linking this ;)

i am wondering whether neural network will work somehow. anyway, i dont know much about that...

Answer 1

[Sean de Wolski]

Perhaps a point-wise least squares method might be of interest to you?

This paper gives a fairly decent description of it for strain calculation in 2-dimensional images - numerical derivatives of a surface. http://www.sciencedirect.com/science/article/pii/S0143816609000189

It gave me good results for my work.