How do I create a Filter that takes out irrelevant data?

Hey everyone,

So I have a large amount of data stored in a matrix that displays the surface of an object when plotted. The only problem is that this contains the whole surface, and not just the roughness curve that I need from it. How can I create a filter (i'm guessing of a gaussian type) that controls the data being output? I don't need code to be provided, just simple direction would be nice.

I had an idea for where to start but I'm not sure if it's right.

A = imread('file_name.txt'); H = fspecial('gaussian',hsize,std)

The thing about this fspecial function is that I do not know what size to specify in the argument. The documentation does not cover this unfortunately.

Thanks, Ian

Answer 1

[Teja Muppirala]

The curve fitting toolbox can help you deal with irregular data. Here I make a noisy surface, approximate it using polynomials, subtract the two, and then display the noise. There is a neat surface fitting GUI tool called "sftool" as well.

x = rand(65000,1);
y = rand(65000,1);
z0 = log(1+x).*sin(3*y);
z = z0 + 0.05*randn(size(z0));

figure
plot3(x,y,z,'k.','markersize',1);
hold on;
F = fit([x y],z,'poly55');
plot(F);
alpha 0.5;

figure
subplot(211)
plot3(x,y,F(x,y) - z,'.','markersize',1);
subplot(212)
plot(F(x,y) - z);

[Ian Wood]

This is exactly what I was looking for. Modifying this code a little for my code, I come up with the profile roughness I need to extrapolate. Thanks Teja, and thanks to you too Ben.

Answer 2

[Ben Mitch]

Hi Ian

There's some information missing from your question, but it sounds like you're looking to interpolate a large amount of data to summarize its form with a small data set.

If so, you might get some mileage out of interp2() and filter2(), if that's the case (interp() and filter() if your "surface" is 1-D). Use interp2() to get the value of your function on a high-resolution regular grid, then use filter2() to smooth it as much as you like (use a matrix of ones as "b" in filter2()), then use interp2() again (or simply decimate) to obtain a smoothed surface on a low-resolution regular grid.

If you need more control over the curvature of the summary surface, you could try instead polynomial estimation. Denote your input data [x, y] as X, and your output data [z] as Y, and use polyfit() to fit an nth-degree polynomial of your choice, then use polyval() to provide a summary data set on a regular grid.

Cheers

[Ian Wood]

Hi Mitch,

Thanks for the reply. What you said in the first paragraph is the idea yes. I will see what I can come up with using your suggestion and let you know how it goes.

Regards

Answer 3

[Ian Wood]

Ok so if I'm reading this correctly the pre-defined function interp2 requires a monotonic matrix, and unfortunately that's not what i have. I will provide more information:

So it's a ~65000 by 3 matrix, that holds x, y, and z plot points to approximate a surface. The problem is it's a surface and not a 2-D roughness curve, which i need to calculate certain parameters, but that code is already done. I just need to know the best method to get this roughness curve.

It should look like a zig-zag pattern. The useful part of the filtering is eliminating the data that represents the surface. Once I have this curve I can take the difference between the two plots (original and filtered), and this will give me the variables i need to implement in my equations.

[Walter Roberson]

What is a "roughness curve" ?

[Ian Wood]

It's terminology for a surface's actual profile at a microscopic level. The heights deviating from the average profile are usually very small (micrometres).