Find light grey circle in greyscale image

Dear community,

I have a problem finding a circle in a greyscale image.

The image was taken with an X-ray scanner, through a circular shaped aperture and contains a ball/circle, approximately in the center of this aperture. I want to find the aperture circle and the (smaller) circle for the ball. I know the approximate radius, so I am specifying a range of radii to look for (since I have other circles in the image that I do not want to find).

The reason I am doing this is to exactly find the centers of these circles in order to evaluate whether the ball was placed at the exact center of the imager. I have a working code that does so, see example below.

Now to the problem:

I now have other images in which the images ball is of a different material and hence not as dark in the image. Unfortunately, I can't find a way to make my current code work for this "light grey" circle.

info = dicominfo('Gantry3-003.dcm');

img = dicomread(info);

img_BW = imbinarize(img,'adaptive','ForegroundPolarity','dark','Sensitivity',1);

SID = info.RTImageSID;

SSD = info.RadiationMachineSAD;

pxlsize_mm = info.ImagePlanePixelSpacing(1) /SID *SSD;

angle_Gantry = round(info.GantryAngle);

angle_Colli = round(info.BeamLimitingDeviceAngle);

angle_Table = round(info.PatientSupportAngle);

h = figure('Position', [2000, 100, 1400, 1000]);

subplot(1,2,1)

imshow(img,[]);

xlim([320,960]);

ylim([320,960]);

subplot(1,2,2)

imshow(img_BW);

xlim([320,960]);

ylim([320,960]);

[centersDark,radiiDark] = imfindcircles(img_BW,[10 20], ...

'ObjectPolarity','dark', 'Sensitivity',0.8);

[centersBright,radiiBright] = imfindcircles(img_BW,[20 30], ...

'ObjectPolarity','bright','Sensitivity',0.8);

numcirclesDark = length(centersDark)/2;

numcirclesBright = length(centersBright)/2;

hDark = viscircles(centersDark,radiiDark);

hBright = viscircles(centersBright, radiiBright,'Color','b');

This is the result (original image on the left, processed image with found circles on the right):

For the new image with a lighter colored ball in the center, it looks like this - I don't find this circle:

Unfortunately, I cannot attach dicom images, so I attached the images given above.

Any help would be greatly appreciated!

Also: If there is a simpler or better solution for my problem in general - feel free to comment! ;)

Thank you very much,

Benedikt

[Benedikt Thomann]

I forgot to mention:

I realize that my problem stems from binarizing the image. I am binarizing the image so that it is easier to find the circles. If there is any other/better way, any help would be welcome!

Answer 1

[Stephan]

Maybe this is an approach:

img = imread('BinarizedImage_NewImage.JPG');
img2 = img;
img2(img2<240) = 0;
imshowpair(img,img2,'montage')

result is attached. The value 240 appears to work good - 235 gives similar results. You should play with the values. Once you have a suitable value, you can use imfindcircles as always.

[Benedikt Thomann]

Sorry for the delayed reply - was absent for a couple of days. Suggested code seems to work.

For reasons I do not quite understand, it works with a value of 1450 in my case (pretty much does nothing with 240) but I'll take it! :) Maybe it is because I import the original dicom file and extract the image from there.

Thank you very much for your help!

[Benedikt Thomann]

If I may, I have a follow-up issue.

The circle detection in itself works but the inner circle appears to be more off-center than in the original picture. See image attached.

So the code seems to overestimate the shift of the inner circle from the center of the outer circle.

Any idea?

[Stephan]

i think you can only try to play around with the treshold and try to find the best value for your needs.

[Benedikt Thomann]

Will do, thanks again!

[Image Analyst]

It doesn't look like your images were cropped from anywhere near the same area. But anyway, using one or two thresholds you should be able to find both the bright donut and the darker inner spot. Then you can ask regionprops to find the centroid of each blob. Knowing that you can determine the shift between the outer and inner one.

I would do something different than what @Stephan did. Let me know if you want to see it. If so, attach your good (aligned) image and your bad (misaligned) image.

[Benedikt Thomann]

Thank you for your comment. I think my last question was a little bit misleading.

I am not concerned about the two images in my last comment not being perfectly aligned. That was merely a screenshot from the plot I am generating in my script to see the original and the edited image. I do not know why they are not perfectly aligned but I do not see any real problem in that either as all I want to do is compare the two centers of the found circles relative to each other.

My real problem was that to me it looks as if the inner black circle was somehow further away from the center than the light grey circle in the original image. But maybe it's just my eyes or some kind of illusion. I don't see any reason why it should "move". Tha data shouldn't lie. :)

Despite of that: I am indeed interested in the double thresholding solution you mentioned to maybe increase accuracy in terms of circles found.

Unfortunately, I cannot add dicom files, or can I?

[Stephan]

@Image Analyst Im also interested in an alternative - probably more exact solution - to this.