Major axis and minor axis

How can I plot the major axis and minor axis for an object in an image?

[Pamela]

I need your held please !!

[Image Analyst]

I've just added a section on "How do I create an ellipse" to the FAQ: http://matlab.wikia.com/wiki/FAQ#How_do_I_create_an_ellipse.3F so that might help you.

Answer 1

[Walter Roberson]

regionprops() to get the centroid and orientation (angle). After that it becomes straight trig. The axes length is R, compute R*cos(theta), R*sin(theta), and center those displacements on the centroid.

[Teja Muppirala]

For example:

http://blogs.mathworks.com/steve/2010/07/30/visualizing-regionprops-ellipse-measurements/

[Pamela]

i tried this. Can you help me to complete it please??

img_seg=segmentation(image);
a = regionprops(img_seg, 'Orientation', 'MajorAxisLength', 'MinorAxisLength', 'Centroid');
theta = pi*a.Orientation/180;
x=a.MajorAxisLength*cos(theta);
y=a.MinorAxisLength*sin(theta);

[Pamela]

How can i excute this with an image (black background and a color region of interest)

[Image Analyst]

I think Steve's blog that Teja referred you to spells it out pretty explicitly. Can't you follow that? Do you want something different than that, like lines instead of ellipses?

[Pamela]

I tried it but i had this error

??? Error using ==> iptcheckinput Function REGIONPROPS expected its first input, L, to be integer-valued.

[Image Analyst]

You must have a very old version of MATLAB. Try labeling your binary image before passing it in, instead of passing in just the binary image:

labeledImage = bwlabel(img_seg);
measurements = regionprops(labeledImage, 'Orientation', 'MajorAxisLength', 'MinorAxisLength', 'Centroid');

[Pamela]

thx it works but if i want to draw the major axes and the minor axes and to hide the ellipse, what should i do??

[Image Analyst]

Use the centroid, orientation, and the two lengths and some simple 10th grade trigonometry (see Walter's response) to find the endpoints of the lines and then use line() to draw the major and minor axes. The angle for the minor axes will be the major axis angle + 90 degrees. There are degree versions of sin and cos called sind() and cosd() if you prefer to work in degrees instead of radians. IMPORTANT NOTE: the axes are determined by an elliptical fit so don't expect the endpoints of the axes to exactly hit the boundary of the object.

[Pamela]

I tried this code. Can you correct it for me and how can i use this axes and PCA to study the asymmetry of the object.

xfact = s.MajorAxisLength./2*cos(s.Orientation*pi/180);
yfact = s.MajorAxisLength./2*sin(s.Orientation*pi/180);
Major = line(xfact+s.Centroid(1),yfact+s.Centroid(2));

[Image Analyst]

Pamela, I can't run your code because I don't have your image or your segmentation function. But common reasons for things being reversed are getting (row, column) confused with (x,y) when it should correspond to (y, x), and/or plotting points on a different axes than the image, where the YDir property is going the opposite way. Recall that the standard for images is for the top line to be 1 and higher line numbers are below that. This is opposite to the convention that you have is you just use plot() to plot some numbers in a brand new axes. However, plot will be correct if you used "hold on" and plotted in the same axes right over the top of your image.

[Pamela]

Thank you, It works now. I managed to draw the two axes but now I need to study the asymmetry of the region. Can you tell me how can I use these axes and the PCA to study the asymmetry of the ROI.

[Image Analyst]

Can you do something along these lines:

positivePC1Elements = PC1_image >= 0;
negativePC1Elements = PC1_image < 0;
meanPositivePC1 = mean(PC1_image(positivePC1Elements));
meanNegativePC1 = mean(PC1_image(negativePC1Elements));

if abs(meanPositivePC1 - meanNegativePC1) < someTolerance
  % They're close
else
  % They're different
end

[Pamela]

Thank you Image Analyst, Can you explain to me more because I have no idea about PCA. What represents the variables used??

[Pamela]

Bonjour Et comment faire pour ajouter 2 axes de plus et pour pouvoir diviser la région en 8 parties en utilisant 4 axes en total??? Je compte sur votre aide pour pouvoir avancer

[Image Analyst]

I can't really give you a course in PCA here. Basically they are the major directions of your independent variables, which can be combinations of your actual variables.

[Pamela]

Hi! To study the similarity between 2 images we should use these steps???

-Calculate the mean of the input images

-Subtract the mean from the input images to obtain the mean-shifted images

-Calculate the eigenvectors and eigenvalues of the mean-shifted images

-Order the eigenvectors by their corresponding eigenvalues, in decreasing order

-Retain only the eigenvectors with the largest eigenvalues (the principal components)

-Project the mean-shifted images into the eigenspace using the retained eigenvectors

-Calculate the Euclidean distance

It's the same solution to study the asymmetry of the region?? Can this help me??

Thanks

[Image Analyst]

Pamela, I haven't heard of that method to determine similarity. Much more common is SSIM ( http://en.wikipedia.org/wiki/SSIM) and friends, or even PSNR. Or you can use image moments: http://en.wikipedia.org/wiki/Image_moment