0
$\begingroup$

enter image description here I am trying to fit a second order polynomial to various curves on a binary image using MATLAB polyfit function.There are many such curves on the image, and the coordinates of the pixels forming the curve are known.

How do I compare the co-efficients, so as to arrive at one approximation to all such curves?

Here are some co-efficients of such curves:

 0.0010126     -1.5981       1074.8
 0.0024871     -3.3327       1866.3
 0.002662      -3.7067       1631.4
 0.0025574     -4.5129       2148.4
 0.002114      -3.3457       2195.6
 0.0015383     -2.6023       1682.9
Improve this question
$\endgroup$
  • $\begingroup$ Can you use polyfit on a all the data at once? $\endgroup$ – Pier-Yves Lessard Oct 21 '18 at 2:53
  • $\begingroup$ No. These curves are the edges of the ribs on a chest x ray. They are slightly different from each other, but have a characteristic similarity. $\endgroup$ – Prashant Oct 21 '18 at 2:56
  • $\begingroup$ Could you share a sample of data? What is the difference between the curves. You ask how to compare them, but according to what criteria?A little context would help a lot here. $\endgroup$ – Pier-Yves Lessard Oct 21 '18 at 2:56
  • $\begingroup$ I want to detect the rib edges on a chest X ray. Each of the rib edges are different and are not a continuous lines. Some are only with fragments making up the whole rib edge. It is possible to fit polynomial to each segment of the edge. But I want to study on a possibility of a general or approximated polynomial applicable to many if not all of such curves. I have added an image for ref $\endgroup$ – Prashant Oct 21 '18 at 3:05
  • 1
    $\begingroup$ To find a single polynomial, you could generate a new point cloud that would mix all of your polynomial, limiting them to a x that has a meaning (short rib generates less points than long ribs, kind of..), then rerun polyfit ont hat new cloud. Or, you could fit a curve to find the polynomial coefficient. For example, you gave 6 polynomials representing a rib each. You could make 3 polynomials (a0,a1,a2) that varies relatives to the ribs position. Actually, I tried to plot your coefficients (taken vertically) and they somewhat looks like a parabola, maybe the second approach may work. $\endgroup$ – Pier-Yves Lessard Oct 21 '18 at 3:10
0
$\begingroup$

Based on what you share, this is the best answer I can give. I see two methods:

Method #1,

Instead of making N polynomials (if N is the number of ribs), you could fit a curve for all the ribs at once. Of course, for that, you will need to have a common (0,0) coordinate, in other word, stack them one over the other on your image. Since you are able to make a distinction between the ribs, that means that you can already isolate them, so that shouldn't be an issue.

Or if you prefer, you can generate a new cloud of points with the polynomials you already found. 

coeffs = [0.0010126, -1.5981, 1074.8; 0.0024871, -3.3327, 1866.3];
x = linspace(0,1000,1000); % fictives points
merged_x = [x x];
merged_y = [ polyval(coeffs(1,:), x), polyval(coeffs(2,:),x)];
coeffs_final = polyfit(merged_x, merged_y, 2);
figure
hold on
plot(polyval(coeffs(1,:), x), 'red')
plot(polyval(coeffs(2,:), x), 'blue')
plot(polyval(coeffs_final, x), 'black')

enter image description here

Method #2

You could fit 3 polynomials to find the values of the coefficient you already have found. If we take the first polynomial you provided, then $a_2=0.0010126$ $a_1=-1.5981$ $a_0=1074.8$

You could fit a curve that would follow all $a_2$, another for all $a_1$, and same for all $a_0$. This would yield 9 new coefficients (3 polynomials) that will allow you to find your polynomials based on the rib position.

Here's an example

coeffs = [ 0.0010126     -1.5981       1074.8;
    0.0024871     -3.3327       1866.3;
 0.002662      -3.7067       1631.4;
 0.0025574     -4.5129       2148.4;
 0.002114      -3.3457       2195.6;
 0.0015383     -2.6023       1682.9 ];

x = linspace(0,1000,1000); %fictive value
ribs_count = size(coeffs, 1);
a2 = polyfit(1:ribs_count, coeffs(:,1)', 2); % careful to the ' symbol that transpose the column into a row
a1 = polyfit(1:ribs_count, coeffs(:,2)', 2);
a0 = polyfit(1:ribs_count, coeffs(:,3)', 2);

figure
for rib=1:ribs_count
    subplot(6,1,rib);
    hold on
    new_coeffs = [polyval(a2,rib), polyval(a1,rib), polyval(a0,rib)];
    plot(x, polyval(new_coeffs, x), 'red')
    plot(x, polyval(coeffs(rib,:), x), 'blue')
end

enter image description here

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.