MATLAB Answers

bobby
1

Surface area from a z-matrix

Asked by bobby
on 15 Apr 2012
I have x, y axis values and corresponding z values. This allows me to plot the shape using surf function. Example:-
How to calculate surface area of this irregular shape?
Please i'm a novice learning this stuff, i have come along quite well since last week or two. Thank you for all your help. The other answers provided before for irregular shapes have me confused, since i don't know how to triangulate.
Additionally i also want to calculate surface area above a certain height. So i'll cut all the values below a certain height and then possibly calculate surface area above a cut-off height.
Edit:
Thanks to inputs by the community and basically whole programme written by Richard. Here's the code.
function sa(Z, cutoff)
%Credit: Richard Brown, MATLAB central forum (http://www.mathworks.com/matlabcentral/answers/)
dx=0.092; % x-axis calibration
dy=0.095; % y-axis calibration
[m, n] = size(Z);
areas = 0.5*sqrt((dx*dy)^2 + (dx*(Z(1:m-1,2:n) - Z(1:m-1,1:n-1))).^2 + ...
(dy*(Z(2:m,1:n-1) - Z(1:m-1,1:n-1))).^2) + ...
0.5*sqrt((dx*dy)^2 + (dx*(Z(1:m-1,2:n) - Z(2:m,2:n))).^2 + ...
(dy*(Z(2:m,1:n-1) - Z(2:m,2:n))).^2);
zMean = 0.25 * (Z(1:m-1,1:n-1) + Z(1:m-1,2:n) + Z(2:m,1:n-1) + Z(2:m,2:n));
areas(zMean <= cutoff) = 0;
surfaceArea = sum(areas(:));
sprintf('Total surface area is %2.4f\n', surfaceArea)
return
end

  1 Comment

I've just had it pointed out to me that there is a small mistake in this code. Assuming that x corresponds to columns, and y to rows, then the first two lines (but not the second two lines) of the areas calculation has dx and dy backwards. I've fixed it in my original answer below.

Sign in to comment.

4 Answers

Answer by Richard Brown on 16 Apr 2012
Edited by Richard Brown on 15 Jul 2013
 Accepted Answer

OK, well how about splitting each quadrilateral into two triangles, and just summing up the areas? I'm sorry there's no way I can make this look pleasant, but ...
[m, n] = size(Z);
areas = 0.5*sqrt((dx*dy)^2 + (dy*(Z(1:m-1,2:n) - Z(1:m-1,1:n-1))).^2 + ...
(dx*(Z(2:m,1:n-1 - Z(1:m-1,1:n-1)))).^2) + ...
0.5*sqrt((dx*dy)^2 + (dx*(Z(1:m-1,2:n) - Z(2:m,2:n))).^2 + ...
(dy*(Z(2:m,1:n-1) - Z(2:m,2:n))).^2)
surfaceArea = sum(areas(:))
edit: Jul 15, 2013 There was a mistake, and dx and dy were backwards in the first two lines of the code. The code has been corrected now.

  3 Comments

bobby
on 16 Apr 2012
That is one awesome code Richard. I can't thank you enough. Here i was searching for triangulations and TINs since last thursday. It gives me result within 3% tolerance level, which is absolutely awesome. Just a small correction in the code, it is missing a bracket.
areas = 0.5*sqrt((dx*dy)^2 + (dx*(Z(1:m-1,2:n) - Z(1:m-1,1:n-1))).^2 + ...
(dy*(Z(2:m,1:n-1) - Z(1:m-1,1:n-1))).^2) + ...
0.5*sqrt((dx*dy)^2 + (dx*(Z(1:m-1,2:n) - Z(2:m,2:n))).^2 + ...
(dy*(Z(2:m,1:n-1) - Z(2:m,2:n))).^2);
surfaceArea = sum(areas(:))
bobby
on 16 Apr 2012
Now how to find the surface area above a cut-off height. The present code connects the empty valleys and assumes it to a flat surface. I want to exclude them. Any intelligent ideas?
For each of the areas in the 'areas' array, you could work out the mean Z value
zMean = 0.25 * (Z(1:m-1,1:n-1) + Z(1:m-1,2:n) + Z(2:m,1:n-1) + Z(2:m,2:n))
and then
areas(zMean > cutoff) = 0;
surfaceArea = sum(areas(:));
It's a bit sloppy - you should probably do this for the individual triangles, but it will certainly give you a reasonable approximation

Sign in to comment.


Answer by Walter Roberson
on 15 Apr 2012

I do not know if you will be able to calculate the surface area of the regular shape interspersed with the irregular tops. When I look at the image with the irregular tops, it looks to me as if the tops are fairly irregular, possibly even fractal. I don't know if a meaningful surface area could be calculated: the surface area of a fractal is infinite.
For the first figure, I can think of a crude way to find the area, but I think there are simpler ways. I would need to think further about good ways to find the area. But to cross-check: are your x and y regularly spaced, or irregularly ?

  7 Comments

bobby
on 16 Apr 2012
hahaha Sean, i don't know if you are wanting me to work out the approach or commenting on Richard's script. I meant that i'm getting 20% error with Richard's script, which i think is significant. I'm in no way being facetious and putting down gentlemen giving me suggestions.
Sean de Wolski
on 16 Apr 2012
I wasn't commenting on Richard's script at all, but rather the fractal nature of surface area (and length) of irregular shapes. I apologize if that came across the wrong way.
http://www.google.com/#hl=en&gs_nf=1&cp=19&gs_id=1k&xhr=t&q=how+long+is+the+coast+of+britain&fp=46b06bf633a62cc4
bobby
on 16 Apr 2012
Ah, the good old mandelbrot. Yes, I understand your point now. That is why i edited my post as i did understood that it will be nearly impossible. But the surface area of the surface that i'm after, given in the post, i'm sure a solution exists. Thanks to you i came across a very neat paper by Mandelbrot.

Sign in to comment.


Answer by Richard Brown on 16 Apr 2012

If your data is smooth enough (assuming that's what you're after), then there is a really quick way to work out approximately the surface area, using the normals that Matlab computes when plotting the surface. To make it simple I'll assume you have a uniform grid with spacings dx and dy.
dA = dx * dy;
h = surf( ... )
N = get(h, 'VertexNormals');
N_z = squeeze(N(:, :, 3));
% Normalise it
N_z = N_z ./ sqrt(sum(N.^2, 3));
Area = dA * sum(1./N_z(:));

  3 Comments

You *need* to smooth this first though, and the solution will only be approximate (probably to about 2sf). Otherwise you'd need to calculate the surface area of an interpolant, which would be a bit more effort intensive
bobby
on 16 Apr 2012
Thank you Richard for your interest and guidance. That is a very neat way of getting surface area. Yes, the grids are uniformly spaced. I am after a quite accurate solution here, since between different surfaces there can be a very small difference and i need to know that. This script is way off right now, way more than 2sigma. So i was looking at some TIN algorithms, i will create triangulated meshes and find individual triangle area. But even after spending weekend on this, i am nearly nowhere.
Hence my comment about smoothing - if you apply that straight to the mesh you've got there, all bets are off! Anyway, see my next answer, it might be more in line with what you're after.

Sign in to comment.


Answer by bobby
on 16 Apr 2012

function surfacearea(Z, cutoff)
dx=0.092;
dy=0.095;
[nrow, ncol] = size(Z);
for m=1:nrow
for n=1:ncol
zMean = mean(Z(1:m-1,1:n-1) + Z(1:m-1,2:n) + Z(2:m,1:n-1) + Z(2:m,2:n));
if(zMean >= cutoff)
areas = 0.5*sqrt((dx*dy)^2 + (dx*(Z(1:m-1,2:n) - Z(1:m-1,1:n-1))).^2 + ...
(dy*(Z(2:m,1:n-1) - Z(1:m-1,1:n-1))).^2) + ...
0.5*sqrt((dx*dy)^2 + (dx*(Z(1:m-1,2:n) - Z(2:m,2:n))).^2 + ...
(dy*(Z(2:m,1:n-1) - Z(2:m,2:n))).^2);
end
end
end
surfaceArea = sum(areas(:));
return
end
This is how the final code looks like with a cut-off limit included. Any inputs into speeding up this code? I have 1024*1280 matrix.

  3 Comments

I think you may have slightly misunderstood some of my earlier answers - you don't need to have any loops at all. All you should need to do is:
areas = 0.5*sqrt((dx*dy)^2 + (dx*(Z(1:m-1,2:n) - Z(1:m-1,1:n-1))).^2 + ...
(dy*(Z(2:m,1:n-1) - Z(1:m-1,1:n-1))).^2) + ...
0.5*sqrt((dx*dy)^2 + (dx*(Z(1:m-1,2:n) - Z(2:m,2:n))).^2 + ...
(dy*(Z(2:m,1:n-1) - Z(2:m,2:n))).^2);
zMean = 0.25 * (Z(1:m-1,1:n-1) + Z(1:m-1,2:n) + Z(2:m,1:n-1) + Z(2:m,2:n))
areas(zMean > cutoff) = 0;
surfaceArea = sum(areas(:));
And that should run pretty fast
bobby
on 17 Apr 2012
Oh yes. This does not require loops. Let me run it again. Thanks Richard.
bobby
on 17 Apr 2012
Yes, the code runs fast, only thing is there should be < sign. Thanks a lot Richard for helping me through this ordeal. I'll definitely shop again. :)

Sign in to comment.