To detect circular objects in a binary image with Matlab, I would use 'regionprops'.

Use both 'area' and 'perimeter' to calculate the radius of each (would be) circle. The more those two radii are similar, the more circular the blob is.

A = pi * r * r

P = 2 * pi * r

However, the blob you indicated will not be detected this way because there are blobs in the image that are more circular! So I agree with Adam: you need to define what sets the indicated blob apart from the other ones. In this image you could use that it is the largest closed contour without a hole, but does the same hold for other images?

As to eigenvalues: is your blob always oriented in the same way? If the blob is really circular, it has no principal components!

You may use Hough transform for circles.

wiki link: http://en.wikipedia.org/wiki/Hough_transform

Matlab link: http://www.mathworks.com/matlabcentral/fileexchange/26978-hough-transform-for-circles

Dear Nisreen

You can use many techniques like moments ,chain codes,... etc

You can find all features in chapter 7 on this link:

http://www.itu.dk/courses/SIGB/F2011/untitled%20folder/Reading/RelatedBooks/FeatureExtraction&Imageprocessing.pdf

I see several circles in this image, what features of the blob that has the arrow set it apart from the other circles in the image?

Dear Nisreen,

If your concern is only circular objects then you can use convex hull also like the Hough transform. The convex hull is useful if the object are not a proper circle this would help you to identify those objects also if required.

The wiki link for convex hull is: http://en.wikipedia.org/wiki/Convex_hull

The matlab link is: http://www.mathworks.com/matlabcentral/fileexchange/40520-convex-hull-implementation-matlab-code

Regards,

Chayan Halder

Dear Nisreen,

You can use the Hough transform for detection of cicular-like objects but it has high computing time.

To detect circular objects in a binary image with Matlab, I would use 'regionprops'.

Use both 'area' and 'perimeter' to calculate the radius of each (would be) circle. The more those two radii are similar, the more circular the blob is.

A = pi * r * r

P = 2 * pi * r

However, the blob you indicated will not be detected this way because there are blobs in the image that are more circular! So I agree with Adam: you need to define what sets the indicated blob apart from the other ones. In this image you could use that it is the largest closed contour without a hole, but does the same hold for other images?

As to eigenvalues: is your blob always oriented in the same way? If the blob is really circular, it has no principal components!

I used region descriptors: area, degree of circularity, perimeter, solidity, ... EulerNumber.

I am trying to find "standard features" according to aneurysms characteristics ... 

1- Detecting circular objects in an image can be done easily using Hough Transform, as pointed earlier,

BUT you should do one step before getting the HT, which is finding the edges or the outer contours in the image.

Here is a link for a tutorial for HT and its Matlab code:

https://www.cis.rit.edu/class/simg782/lectures/lecture_10/lec782_05_10.pdf

2- For a standard features for a circular objects you can use Fourier descriptors which are invariant under naive transformations.

Good luck!

I really appreciate all researchers who gave an answer

I want to detect cerebral sacular aneurysms in 2D DSA angiogram  images ... they are not really circular objects ... (Reign Of Interest(ROI) referd to by red arrow). Actually they have not regular shape.

An important thing is that we don't know diameter or center of circular-like( or blob ) objects.

Any additional invariant features ?? And algorithms to detect and recognize ROI?

For your problem you can see these links:

http://imageprocessingblog.com/detecting-and-counting-objects-with-circular-features/

here multiple circular type objects are detected. Also you can see this link:

http://in.mathworks.com/help/images/examples/identifying-round-objects.html

For a good algorithm for circular object identification you can read the paper: J. Chen, C. Hu, X. Yuan, Z. Feng, H. Miao, "An identification and classification method for circular object based on rotating image template matching", Mechatronics and Automation (ICMA), pp. 1338 - 1343, 2013

Hi,

Please give us more info about your objects. If you have any size constraints and every time binary objects, maybe morphological filtering with specific size for the structure element and considering a circular shape for that can be helpful. Another thing coming in my mind for circular binary objects is 'dist' function in Matlab. Please check the example provided for Watershed Segmentation in Matlab, it can be helpful.

I suspect these 'blobs' don't have a particularly distinctive size or shape compared to say, a natural kink in a blood vessel viewed from the wrong angle in 3D. So I'd be tempted towards algorithms based around the distance transform that look for sudden changes in the local width of the blood vessels, because this closely models the underlying nature of an aneurysm. You could apply ridge detection to the distance transform first (this can be a bit hit-and-miss) in an attempt to track the path of the blood vessels and then look for local changes in width distribution along the path of the vessel. Perhaps a window function might be sufficient, where you train on attributes that examine the shape and modality of the distribution of distances within the window. I'd expect to see changes in the modality/shape of the distribution near an aneurysm.

Thank you all,

I have used rule based classification to solve this problem.

 Hi Nisreen Sulayman

how did you solve your problem?

if you find some codes pls can you pass it to me?

with my thanks

Hi

To detect circular objects in an image you can use Circular Hough Transform (CHT) algorithm.

This link may help you to find out more about it:

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

If you are using matlab, just use function regionsprop to extract all necessary features https://www.mathworks.com/help/images/ref/regionprops.html