How can I find the round polygons

Hello everybody!Wish you have a good day!

I want to automatically find the round polygons (it is the binary information of myocardium) in these images in order to locote and get their centers. But "imfindcircles" won't find it. Could you please tell me how to find these polygons? The following is the images and what I want to get (I marked the parts I wanted manually) The 3 orignal images:

The images I want:

Thank you for helping!

Best regards Neil

[Michael Haderlein]

In the first image, what qualifies the marked object to be the circle instead of the object south east of it? Same in the third image, why not the object north east of the marked object?

[Neil Liu]

@Michael Haderlein Hi! Actually it is the myocardium of left ventricle after binary processing. The marked one is the contour of myocardium.

[John D'Errico]

Michael makes a very valid point. There are multiple regions that would be as easily considered "round" as the one you chose. You need to explain what criterion makes you choose one over the other.

[Neil Liu]

@John D'Errico, dear sir, this is the binary image of the left ventricle. My marked object is the myocardium. I want to mark it and then get its center.

[Mohammad Abouali]

for me the other object that you haven't marked looks more round.

So, just being round seems not enough to distinguish between myoc* and not a myopic*.

So what are the other criteria to classify a patch as myocardium. (1) being round, (2) ....

[Neil Liu]

@Mohammad Abouali, dear sir, this is the point. I have no idea how to get the criteria. In fact, I just need to know the center of the object I marked.

[Michael Haderlein]

Ok, but only given these processed images, how do you as an expert know that the object of interest are the ones you have marked an not the other ones? We cannot find the criterion because we have no idea of myocardia and I guess most people here would have chosen the ones I have described in my first comment. A criterion could also be that these nearly perfect octagons are not the correct objects. Then you can first filter them out and then find the roundest element. But if then you have a fourth image and there is a hexagon which isn't what we're looking for, this will mess up everything again. So we first need a good criterion.

Answer 1

[Image Analyst]

This is really not very hard. Circularity is the usual method that works pretty well in most cases.

circularities = perimeters .^ 2 ./ (4 * pi * areas); % <= ~3 or 4 is pretty rounded

There are other criteria you can use if you need to: http://en.wikipedia.org/wiki/Roundness_%28object%29. Let me know if you still need help and I can whip out a program in a few minutes.

[Neil Liu]

Dear sir, as Mr.Michael Haderlein's comments "what qualifies the marked object to be the circle instead of the object south east of it? Same in the third image, why not the object north east of the marked oject?" Could I use this method to find the center blob?

[Image Analyst]

Well they're certainly close. You'd have to look at a bunch of properties and see exactly which values describe the one you want and not the others. You might have to have very close/tight thresholds. You're the one who picked that particular blob so I would not think you would ask - I think you would know what's different about that one. Maybe your criteria is to pick only one blob, and you pick the one blob that is round enough and is closest to the center of the image. Who knows? I don't know - it's your image and your decision. You should know.

[Neil Liu]

Thank you Mr Image Analyst. I know the coordinate range of the blob. So it could be the criteria