How to cluster a binary image?

I have the following image. Now I want to separate the black and white pixel from the image so that I can fit an ellipse to each black pixel area and find the length of the major axis of each ellipse. How can I separate black pixel from the white pixel and fit an ellipse?

Answer 1

[Image Analyst]

See my image processing tutorial: http://www.mathworks.com/matlabcentral/fileexchange/?term=authorid%3A31862

Basically, here are the steps.

% Invert your image so the black things are white.
binaryImage = grayImage < 128;
% Call bwlabel
[labeledImage, numberOfRegions] = bwlabel(binaryImage);
% Call regionprops
props = regionprops(labeledImage, 'MajorAxisLength', 'Orientation');

It's all shown in Steve's blog:

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

[sheli whitson]

Thank you Image Analyst.

[Eyup Demir]

@Image Analyst In the image you attached from @Steve Eddins, two white regions in the middle are merged under one ellipse, because there is a one pixel touching. Is there a way to merge them if there was 2 or 3 pixel between them? I want to be able to control some white regions to be merged based on the closest distance between them.

[Walter Roberson]

https://www.mathworks.com/help/images/ref/imclose.html

imclose() with a structuring element as large as the distance you want to merge.

[Eyup Demir]

I think, I found what I was looking for! It is almost-connected-compenent labeling Thank you Steve! https://blogs.mathworks.com/steve/2010/09/07/almost-connected-component-labeling/

[Image Analyst]

Sometimes you can reduce this effect by using 4-connected labeling instead of the default 8-connected labeling.