How to find the diameter of a circle from an image ?

I tried to find the diameter of a circle through an image using two codes

1) regionprops()

2) imdistline()

The diameter in the image is 100mm. The answer differs everytime I change the code. Can anyone tell me where I went wrong ?

-- region props code

% read the image
a = imread('circle6.jpeg');
% convert the image
b = rgb2gray(a);
be = imbinarize(b);
figure(2),imshow(be);
% using regionprops
stats = regionprops('table',be, 'Area','EquivDiameter','Perimeter');
% storing the values in other variables
x = stats.Area;
y = stats.EquivDiameter;
z = stats.Perimeter;
%using area to find the diameter
r = sqrt(x/(4*pi))*0.2645;
%using perimeter
r1 =( z/(4*pi))*0.2645; 
% using equivDiameter
r3 = y*0.2645;

-- imdistline()

a  = imread('circle6.jpeg');
c = rgb2gray(a);
b = imbinarize(c);
imshow(b);

%measuring
h = imdistline(gca);
api =iptgetapi(h);

%
pause();
%
dist =api.getDistance();
u = menu('Choose measuring unit','Pixels','Centim','Meters');
if (u==1)
    fprintf('The length of the object is: %0.2f Pixels\n',dist);
elseif (u==2)
    dist_cm = dist*0.2645;
    fprintf('The length of the object is: %0.2f Centim\n',dist_cm);
else
    dist_m = (dist*0.2645)/100;
    fprintf('The length of the object is: %0.2f Meters\n',dist_m);
end

Answer 1

[Image Analyst]

What you deceptively call y is actually the diameter:

y = stats.EquivDiameter; % y is the diameter.

No need to do anything else. It will not change on each run if the image does not change.

To calibrate, you need to multiply by some spatial calibration factor. If you know the field of view of your image, you can compute it like this

[rows, columns, numberOfColorChannels] = size(a);
pixelsPerMm = imageWidthInMm / columns;
diameter = y * pixelsPerMm

Of course you need to know imageWidthInMm. If you don't, you can use my attached spatial calibration demo to interactively draw a known distance on your image.

[Tasneem Khan]

Sorry for the late reply. Thank you so much for your reply but before I accept your answer I want to know how exactly do I find the imageWidthInMm. That's the only issue I'm facing rn. Rest assured I there is 2cm accuracy problem, but I can deal with that.

[Image Analyst]

You set it equal to the known real world distance

imageWidthInMm = 100; % Known distance of circle in image.

multiply it by the distance in pixels, like I showed you:

pixelsPerMm = imageWidthInMm / columns;

columns is the equivalent circular diameter as gotten from regionprops off a perfectly accurate, undamaged reference circle.

Answer 2

[KSSV]

I = imread("circle6.jpeg") ; 
I1 = rgb2gray(I) ; 

[y,x] = find(I1~=255) ; 
C = [mean(x) mean(y)] ;  % Center of circle 

dx = C(1)-x ; 
dy = C(2)-y ; 
d = sqrt(dx.^2+dy.^2) ; 
r = max(d) ;   % Radius of circle 

th = linspace(0,2*pi) ; 
xc = C(1)+r*cos(th) ; 
yc = C(2)+r*sin(th) ; 

imshow(I) ;
hold on
plot(xc,yc,'r')

[Tasneem Khan]

I got the diameter as 128mm which converts to 12 cm in real world there is an offset of 2 cm. Let me show you the image boudary which it created for the cirle not quite accurate though.