Optimize ellipse to fit shapes in images

Hi,

Given an image with various shapes, I am trying to fit ellipses that match these shapes the best by varying the ellipse parameters a, b and the angle of the ellipse. So far, using a for-loop and trying all possible combination results in code that works very well but is dreadfully slow (script takes more than 20 min per ellipse and I have hundreds of ellipses to fit).

I wonder, therefore, whether MatLab has a good optimization tool for this. Basically, I feed the ellipse-function the image, the x- and y-pixel coordinates and want it to find the optimal ellipse angle and a and b values (given constraints on a and b). My question is, is there a good, robust and fast way of doing this in MatLab?

Answer 1

[Image Analyst]

You're in luck! This is a built-in capability. Take a look at regionprops() in the Image Processing Toolbox. You'll need the MajorAxisLength, MinorAxisLength, and Orientation measurements so be sure to ask regionprops to return those measurements.

[Anders]

I initially used regionprops(), but it works very poorly in my hands for segmentation. My problem is that the shapes overlap and occasionally have small holes, so when I threshold them and use regionprops, it gets more than half of them wrong.

[Image Analyst]

Use imfill() to fill in the holes. See if that will then get you better results.

[Image Analyst]

You aren't working on this same project that several others are, are you? http://www.mathworks.com/matlabcentral/answers/16908-segmentation-of-overlapping-objects