How to detect and separate squares and circles on image ?

Hi guys.As a total beginner in Matlab i couldn't find a way to do this task myself, so my last hope is someone here.I need to detect squares in one image,then separate them(hide everything else in picture)and count them.Same with circles. The picture is shown below.

Answer 1

[Adam Danz]

This should get you started. It uses regionprops to compute the circularity of each object. I've set a subjective threshold of 0.9 where circularity values >0.9 are defined as circles and <=0.9 are not circles. See inline comments for details. Requires Matlab R2019a or later.

% Read image file, convert to binary
file = 'image.png'; % full path is usually better
RGB = imread(file);
mask = imbinarize(rgb2gray(RGB));

% Detect shapes, compute circularity (Assumption: Shapes do not overlap)
stats = regionprops(~mask,{'Circularity','SubarrayIdx'});
circleThreshold = 0.9; % circularity values greater than this are cirlces 

% Copy the colored image twice; eliminate circles and squares
RGBcircles = RGB; 
RGBsquares = RGB; 
for i = 1:numel(stats)
    circularity = stats(i).Circularity; 
    if circularity <= circleThreshold
        % remove object if it's not a circle
        RGBcircles(stats(i).SubarrayIdx{1}, stats(i).SubarrayIdx{2},:) = uint8(255); 
    elseif circularity > circleThreshold
        % remove object if it's a circle
        RGBsquares(stats(i).SubarrayIdx{1}, stats(i).SubarrayIdx{2},:) = uint8(255); 
    end
end

% Count objects
nCircles = sum([stats.Circularity] > circleThreshold);
nSquares = sum([stats.Circularity] <= circleThreshold);

% Show results
figure
tiledlayout(2,2,'TileSpacing','compact','Padding','compact')
nexttile([1,2])
imshow(RGB)
title(sprintf('Original: %d objects', numel(stats)))
axis on

nexttile
imshow(RGBcircles)
title(sprintf('%d circles', nCircles))
axis on

nexttile
imshow(RGBsquares)
title(sprintf('%d squares', nSquares))
axis on

[Amar Mesic]

Thank you so much for your answer,but i have problems running this code.Do i have to change something myself or it is the problem with my matlab?

[Adam Danz]

What version of Matlab are you using?

If you don't know, run ver() and look toward the top of the print-out in the command window.

Also, are you just replacing the image file name and running the rest of the code as-is? What type of image are you using?

[Amar Mesic]

It is R2017a version 9.2.0.538062

[Adam Danz]

The 'circularity' property became available in R2019a so if you want to use this approach, you'll need to update to a more recent release of Matlab. I've udpated my answer to indicate this requirement.

[Amar Mesic]

Oh ,i understand.One more time, thank you

[Image Analyst]

You can use area and perimeter to get circularities:

stats = regionprops(~mask, 'Area', 'Perimeter', 'SubarrayIdx');
allAreas = [stats.Area];
allPerimeters = [stats.Perimeter];
stats.circularities = (4 * pi * allAreas) ./ allPerimeters .^2;

[Adam Danz]

Thanks, @Image Analyst!

Answer 2

[Image Analyst]

See my attached shape recognition demos. Circularity is one way. Also getting distance from centroid and counting peaks is another way. Most ways will breakdown as the shapes get smaller and more pixellated/quantized. And just because a shape has a certain number of vertices does not indicate a certain shape. For example a pentagon and star both have 5 outward vertices.