How to measure circularity??

How to measure circularity ? What is wrong in this code?

CC=bwconncomp(BW); %BW binary image

labeled=labelmatrix(CC);

stats = regionprops(CC, 'all');

stats.circularity = ((stats.Perimeter) .^2 )./ (4*(pi*(stats.Area)))

[Nisreen Sulayman]

what are the all possible features I can use to pick the smallest circular-like component ??

Answer 1

[Image Analyst]

You didn't use brackets. Try it this way

allPerimeters = [stats.Perimeter];
allAreas = [stats.Area];
allCircularities = allPerimeters  .^ 2 ./ (4 * pi* allAreas);

[Nisreen Sulayman]

I want to pick the smallest circular-like component ... but I have this result

[Image Analyst]

Sort based on the area.

[sortedAreas, sortIndexes] = sort(allAreas);

Then find round indexes

roundBlobs = find(allCIrcularities < someValue);

Then find the first number in sortIndexes that is also in roundBlobs.

[Nisreen Sulayman]

What are all possible features to detect this blobs?...

[Nisreen Sulayman]

How can I find the eigenvalues for each component alone??

[Image Analyst]

I have no idea. Maybe you can try this link. Why do you need that?

[John D'Errico]

It is not obvious to me what is meant by circularity on these blobs. Are parts of those snake-like regions, that happen to have a constant curvature, are they circular?

Anyway, circularity is not something that can be accurately determined from eigenvalues. For example, is a square region circular? But if you chose to somehow compute the eigenvalues of a square region, they would be the same. Would that imply a square is circular? Talk about fitting a square peg in a round hole. Suddenly that would be possible! :)

[Nisreen Sulayman]

I want to detect only the cicular-like object(red arrow) in the binary image. What are the suitable features which can be used for this purpose?

How to process each component in the image ?

[Nisreen Sulayman]

@Image Analyst

Could you please complete the code ....

{

Sort based on the area.

[sortedAreas, sortIndexes] = sort(allAreas); Then find round indexes

roundBlobs = find(allCIrcularities < someValue); Then find the first number in sortIndexes that is also in roundBlobs.

}

I have not really catch the idea.

And please another question....

How can I display all the properties that are given by regionprops for each connected component in the image?

[Image Analyst]

All right. I did it all for you. See attached code, which produces the image below.

[Nisreen Sulayman]

thank you very much ... thank you

I need to try it on another binary images to detect Blobs(maybe with more shape features) ... I will tell you the results.

Any mail? Thank you again

[Image Analyst]

You're welcome. My email is not necessary.

[Hossain Md Shakhawat]

How can I find the glcm features for the detected blobs.

Answer 2

[Matt J]

It might also be useful to check whether the Eccentricity and Solidity of a region are above a certain threshold.

[Nisreen Sulayman]

How can I display all the properties that are given by regionprops for each connected component in the image?

[Matt J]

Just display stats(i) to see the properties of the i-th component.

[Nisreen Sulayman]

I used

STATS = regionprops(L, properties)

to know properties for the connected components in the binary image.... and display the results using stats(i).... BUT how can I know when I write for example stats(1), which is the connected component labeld by "1" in the binary image?? and how to display the image with that component alone??

How can I know properties that belong to the blob(red arrow)?

[Matt J]

Use your label matrix, e.g.,

imagesc(labelled==1)

[Image Analyst]

You can look at my code in my answer where the blobs are labeled with their number in the first figure, and then each blob is cropped out to a separate image with the blob number in the title above the blob.

[Nisreen Sulayman]

thank you very much ... thank you

Answer 3

[Nisreen Sulayman]

Hi Mr. Image Analyst,

I am working on the code that you have written "test.m", I want to display more properties of the blobs in the image ... 'MajorAxisLength', 'MinorAxisLength', 'Orientation' ... After that I want to copy these values to a doc file

I have modified the code but it didn't work

[Image Analyst]

Looks okay. It will write to the command window, not a file. To write to Microsoft Word, you'll have to use ActiveX. Explain in more detail what "didn't work" means to you.

[Nisreen Sulayman]

it didn't write correctly ?!

[Image Analyst]

Make it easy for me to help you. Post the entire script and image. I'll wait for it.

[Nisreen Sulayman]

The image is at the following link http://www.mathworks.com/matlabcentral/answers/uploaded_files/22835/im.jpg

I have modified the code "test.m" ....

To display other properties ( 'MajorAxisLength', 'MinorAxisLength', 'Orientation' .)

I have added lines: 124,125,126

modifiy line 152

add line 158

[Image Analyst]

See fixed code, attached. It will find and report the widths of all blobs.

[Nisreen Sulayman]

Thank you Mr. Analyst,

It is ok now. It was not clear to me how to use fprint with that "crazy symbols".

1) Width of the blob means the horizontal axis??

2) How to display values of properties with a field width of tow digits?

3) Can you give me an advice so I learn how to use fprint with symbols?

[Image Analyst]

1. Yes. 2 and 3. In fprintf() and sprintf(), you have format specifiers. When it sees a % symbol, it looks at the numbers and letter after it to decide how to display the variable. f means floating point, d means integer, and s means string. In %T.Lf, L is the number of numbers to the right of the decimal point. and T is the total width of the numbers to the left and right plus the decimal point itself, and any spaced needed to make the whole space for printing be T characters.

[Nisreen Sulayman]

Thank you very much.

How can I get an alert when you "comment" on an answer?

And one last thing please, how to display only the rounded objects in different colors in the original image

http://www.mathworks.com/matlabcentral/answers/uploaded_files/24119/im.jpg

[Image Analyst]

You can calculate the circularities and then use ismember to extract them

labeledImage = bwlabel(binaryImage);
measurements = regionprops(labeledImage, 'Area', 'Perimeter');
allAreas = [measurements.Area];
allPerimeters = [measurements.Perimeter];
circularities = appPerimeters.^2 ./ (4*pi*allAreas);
roundIndexes = find(circularities < 3)
binaryImage = ismember(labeledImage, roundIndexes)>0;
roundMeasurements = regionprops(binaryImage, 'all');

[Nisreen Sulayman]

May be I haven't asked the question correctly, I want to color the round objects in the gray scale image in "red" or "blue" ... like the following image( I have done this using Paint)

[Image Analyst]

First segment the image to get a binary mask of that thing. Then mask it

red = grayImage;
red(mask) = 255;
green = grayImage;
green(mask) = 0;
blue = grayImage;
blue(mask) = 0;
rgbImage = cat(3, red, green, blue);
imshow(rgbImage);

[Nisreen Sulayman]

Thank you very much Mr. Analyst,

I am grateful for your help.

Answer 4

[Nisreen Sulayman]

I have apllied the code. How can I display the round objects in the original images in different color instead of drawing a circle around them???

Answer 5

[Nisreen Sulayman]

How can I index images according to the resulted segmented image? Is there any thing related to build a database using MATLAB?

Any example or link about this?

[Image Analyst]

I don't know what that means. Explain it.

[Nisreen Sulayman]

You can look at the following: https://www.researchgate.net/post/Can_someone_advise_me_about_an_example_of_Content_Based_Image_retrieval

Could you please give me an example how to build database. 3 or 4 images is enough to "database example." And I will increase the size and modify it to meet some requirements.

[Image Analyst]

I don't use the Database Toolbox but you can look into it. Most people here I think just mean a folder of images when they talk about databases. You could analyze them and save the results of them into a mat file if you want. The database should be as big as you need it to be. Maybe 3 or 4 is big enough, but maybe you want more. I don't know what your plans are. Anyway, this sounds like it should be its own question since it's pretty different than your original question. Maybe you'll get more database programmers to answer. I am not a database programmer.

[Nisreen Sulayman]

Ok,

Regarding segmentation results validation,

1- by visual inspection segmentation algorithm has successfully segmented the region of interest.

2- Results have been also revised\judged by radiologists and they insured the success of algorithm

IS THERE any statistical validation for such kind of results to got accuracy of the algorithm??

[Image Analyst]

Yes, there is. You need to create a Receiver Operating Characteristic, or "ROC Curve". Read more about it in Wikipedia: http://en.wikipedia.org/wiki/Receiver_operating_characteristic

[murk hassan memon]

stats.circularity = ((stats.Perimeter) .^2 )./ (4*(pi*(stats.Area))) i am getting an error of this sign ^ as such the error is too many input argument.. is there any other method to find the circularity ? if yes? then kindly help me out

[Matt J]

@murk,

You have multiple objects in "stats" so you need to do something like,

 result = ([stats.Perimeter] .^2 )./ (4*(pi*[stats.Area]))

Answer 6

[Souhardya Roy]

The formula is inverse of what is given.

[yousra zafar]

can you share the code for finding circular objects . i am also working on aneurysm detection. any guidance from your side will he highly appreciated

[Image Analyst]

yousra, check out Image Processing Articles on Aneurysms

Answer 7

[yousra zafar]

i,m unable to run the code text.m attcahed above, getting error : file not found , can you share with me the image C:\Users\Rose\Documents\Temporary\im.jpg