How to close boundary of a circular object in binary image

Thank you all for being here.

I have an image below

the outer boundary is not closed. So it would be a problem when I calculate area of this bubble.

How can I close the boundary of circular object in image above image automatically?

Answer 1

[Matt J]

load Image

[c,r]=imfindcircles(BW,[10,100]);

area=pi*r^2

area = 1.1299e+03

[CW Hsu]

@Matt J Thanks for your help.

Your method helps me calculate diameters of circular object. But how can I draw boundary to close the circular object in image?

I want image like this below. This is created by Microsoft paint manually.

[Matt J]

load Image

[c,r]=imfindcircles(BW,[10,100]);

[y,x]=size(BW);

BW=0.25*BW + (hypot((1:x)-c(1), (1:y)'-c(2))<=r);

imshow(BW,[])

[CW Hsu]

Hi,

Your method works for me. Really helpful. But I have image sequences. And usually there is not only one circular object in other images. How can I do ?

[Matt J]

Use a loop over the different detected circle data c(i,:) and r(i) to repeat the process for many objects.

[CW Hsu]

Thanks for your help. It works successfully

Answer 2

[Image Analyst]

Try this, where mask is your binary image

mask  = bwconvhull(mask, 'union');
props = regionprops(mask, 'Area', 'EquivDiameter');
area = props.Area
diameter = props.EquivDiameter

[CW Hsu]

@Image Analyst Thanks for your advice.

Since I have many images, usually there is not only one circular object in the image. Can I use 'union'? Or maybe 'object' ?

But for the image I show at the top. I try using 'object' . The image becomes image like this.

[Image Analyst]

You can use the 'objects' option but keep in mind that if the curves are too far apart it will consider them as two separate blobs. So two C shapes might end up as two solid D shapes. Only if their convex hulls overlap would they be considered a single blob. And then you might get something like your bottom image. So you'd have to call bwconvhull a second time to get a convex shape.

Another option would be to use dbscan to identify which pixels are close to each other. You specify that closeness distance. So this essentially labels each group of blobs with a single ID number. Then you can use ismember to extract each group one at a time, and then use the 'union' option since all pixels in the image now belong to a single group. Then repeat for all groups in the image. I'm attaching a demo I created of dbscan. See Wikipedia if you don't know what dbscan is.