How to recognize bean shaped object in image?

Hi,

So far i am able to get measurements of an object in image using regionprops() but now i am not able to decide the object is of type "bean-shaped". How do i recognize it?

any type of help will be great.

Answer 1

[Image Analyst]

I'd look for a solidity in a certain range. To be more robust, so it doesn't also get things shaped like asterisks, you can get the convex hull and xor it with the blob and look for blobs (which represent bays/incursions) in a certain size range. Sorry, I don't already have code to do that, you'll have to write it. You can look at the image segmentation tutorial in my file exchange though, http://www.mathworks.com/matlabcentral/fileexchange/?term=authorid%3A31862

[Yogi]

Thanks for reply. i'll get back once i implement your answer. thanks.

[Image Analyst]

Yogi's "Answer" moved here since it's a reply to me, not an answer to his original question:

Hi Image Analyst, Here is my code:

s=regionprops(bw,'all');
for k=1:length(s)
    if(s(k).Circularity >1.00 && s(k).Circularity < 1.09)
         % circular shape
    elseif(s(k).Circularity > 1.09 && (s(k).Solidity > 0.75 && s(k).Solidity < 0.98))
         %bean shape
    end
end

The problem is the bean-shaped objects' circularity varyies, and I am not able to figure out possible maximum value for it. Will only circularity and solidity will suffice, or I need some more filters to figure out if it's a bean shape?

Thanks.

[Image Analyst]

I think you need to say that your circularity is more than some number also or else a bean might get classified as a circle. Why don't you attach your binary image so I/we can try some things?

[Yogi]

Hi Image Analyst, Please find attached images. These are some samples I have.

Thanks

[Image Analyst]

It's not clear which ones are bean-shaped and which are not. Describe what is different between the two classes of binary shapes. They all have a lumpy shape and incursions of various numbers and areas.

[Image Analyst]

To find incursions, you can get the Equivalent Circular Diameter by asking regionprops for EquivDiameter. Then cut that in half to get a smaller circular mask. Then mask each blob with it's small circular mask and sum the inverted image. If there is a black incursion, you'll get a non-zero sum, meaning some black pixels penetrate into the interior of the blob. Is that what you want to do?

[Yogi]

Hi Image Analyst, Thanks for reply. Samples I am having contains objects which are likely bean shaped and not "exactly" bean shaped. Real data expectation is even worst. So I need some sort of range of values of "parameters" which can decide this object is bean shaped or similar/likely a bean.

[Image Analyst]

Correct.

[Lior]

Maybe you can try adding properties such as: eccentricity, major/minor axis and other parameters. You can take a reference image of a ''bean-shaped'' object - apply regionprops to it and analyze the parameters you get. I would create a database of ''bean-shaped'' objects and see if there's any consistency in the properties of those images. Then I'll know what parameters to analyze. I hope I was somewhat clear... :)