identifying shapes in image

Any help with identifying the shapes in this image. they are incorrectly identified due to their rotaion. this code works great on identifying shapes in other pictures but this one it fails.

%% ========================================================================
%  IMAGE PREPROCESSING + WATERSHED SEGMENTATION + RBC SHAPE CLASSIFICATION
%  ========================================================================
% --- Select Image ---
[filename, pathname] = uigetfile( ...
    {'*.jpg;*.jpeg;*.png;*.tif;*.tiff;*.bmp;*.gif', ...
    'Image Files (*.jpg, *.jpeg, *.png, *.tif, *.tiff, *.bmp, *.gif)'; ...
    '*.*','All Files (*.*)'}, ...
    'Select an image');
if isequal(filename,0)
    disp('No file selected.');
    return;
end
imgPath = fullfile(pathname, filename);
img = imread(imgPath);
% Detect grayscale or RGB
isGray = is_image_grayscale(img);
figure('Name', filename, 'NumberTitle', 'off');
%% ========================================================================
%  PATH A — GRAYSCALE IMAGE
% ========================================================================
if isGray
    % Original
    subplot(3,2,1); imshow(img); title('Loaded (Grayscale)');
    
    % Ensure grayscale
    GRAY = im2gray(img);
    subplot(3,2,2); imshow(GRAY); title('Grayscale Image');
    % Dilation (line)
    se_line = strel("line",10,0);
    dil1 = imdilate(GRAY, se_line);
    subplot(3,2,3); imshow(dil1); title('1st Dilation');
    % Dilation (disk)
    se_disk2 = strel("disk",2);
    dil2 = imdilate(dil1, se_disk2);
    subplot(3,2,4); imshow(dil2); title('2nd Dilation');
    % Erode
    se_disk1 = strel("disk",1);
    erode = imerode(dil2, se_disk1);
    subplot(3,2,5); imshow(erode); title('Erosion');
    % Binary
    BW = imbinarize(erode, 25/255);
    
    subplot(3,2,6); imshow(BW); title('Binary (Threshold 25)');
    pause(2);
    % Invert
    BW_inv = ~BW;
    subplot(3,2,6); imshow(BW_inv); title('Inverted Binary');
    
    
    % ------------------ Watershed ------------------
    figure; 
    D = -bwdist(BW);
    subplot(3,2,1); imshow(D, []); title('Distance Transform');
    mask = imextendedmin(D,2);
    D2 = imimposemin(D,mask);
    L = watershed(D2);
    L(BW) = 0;
    subplot(3,2,2); imshow(label2rgb(L,'jet','w','shuffle'));
    title('Watershed Segmentation');
    subplot(3,2,3); imshow(label2rgb(L,'jet',[0.5 0.5 0.5]));
    title('Watershed Transform');
    %% ========================================================================
    %  PATH B — RGB IMAGE
    % ========================================================================
else
    subplot(3,2,1); imshow(img); title('Loaded (Colour)');
    % HSV saturation adjustment
    HSV = rgb2hsv(img);
    S = HSV(:,:,2);
    gamma = 3;
    S_tweaked = S.^gamma;
    subplot(3,2,2); imshow(S_tweaked, []); title('Saturation (Gamma=3)');
    % Convert to grayscale
    GRAY = im2gray(S_tweaked);
    subplot(3,2,3); imshow(GRAY); title('Grayscale');
    % Morphological closing
    se_close = strel("disk",10);
    close_img = imclose(GRAY, se_close);
    subplot(3,2,4); imshow(close_img); title('Close');
    % Opening
    se_open = strel("disk",3);
    open_img = imopen(close_img, se_open);
    subplot(3,2,5); imshow(open_img); title('Open');
    % Binary image
    BW = imbinarize(open_img);
    subplot(3,2,6); imshow(BW); title('Binary');
    pause(2);
    BW_inv = ~BW;
    subplot(3,2,6); imshow(BW_inv); title('Inverted Binary');
    % ------------------ Watershed ------------------
    figure;
    D = -bwdist(~BW);
    subplot(3,2,1); imshow(D,[]); title('Distance Transform');
    mask = imextendedmin(D,2);
    D2 = imimposemin(D,mask);
    L = watershed(D2);
    L(~BW) = 0;
    subplot(3,2,2); imshow(label2rgb(L,'jet','w','shuffle'));
    title('Watershed Segmentation');
    subplot(3,2,3); imshow(label2rgb(L,'jet',[0.5 0.5 0.5]));
    title('Watershed Transform');
end
%% ========================================================================
%  SHAPE CLASSIFICATION (CIRCLE / TRIANGLE / SQUARE / RECTANGLE)
% ========================================================================
figure; imshow(img); hold on;
labels = unique(L);
labels(labels==0) = [];   % remove background
for k = 1:length(labels)
    regionMask = (L == labels(k));
    % Stats
    stats = regionprops(regionMask,'Area','Perimeter','BoundingBox','Centroid');
    A = stats.Area;
    P = stats.Perimeter;
    BB = stats.BoundingBox;
    C = stats.Centroid;
    % RBC classification
    shape = classifyPrimitive(A, P, BB);
    % Put label on image
    text(C(1), C(2), shape, ...
        'Color', 'white', ...
        'FontSize', 12, ...
        'FontWeight', 'bold', ...
        'HorizontalAlignment', 'center');
end
hold off;
%% ========================================================================
%  FUNCTIONS
% ========================================================================
function tf = is_image_grayscale(I)
if ndims(I)==2
    tf = true;
elseif ndims(I)==3 && size(I,3)==3
    tf = isequal(I(:,:,1),I(:,:,2)) && isequal(I(:,:,1),I(:,:,3));
else
    tf = false;
end
end
% --- RBC Primitive Classifier ---
function label = classifyPrimitive(area, perimeter, bbox)
w = bbox(3);
h = bbox(4);
%% Circle metric from RBC paper (should be ~1)
circle_metric = (perimeter^2) / (4*pi*area);
if abs(circle_metric - 1) < 0.158
    label = 'Circle';
    return;
end
%% Triangle metric: area approx half bounding box
bbox_area = w*h;
area_ratio = area / bbox_area;
if area_ratio > 0.30 && area_ratio < 0.65
    label = 'Triangle';
    return;
end
%% Square vs Rectangle
aspect_ratio = w/h;
if abs(aspect_ratio - 1) < 0.60
    label = 'Square';
else
    label = 'Rectangle';
end

Answer 1

[Matt J]

See this FEX download,

https://www.mathworks.com/matlabcentral/fileexchange/181422-fit-polyshape-to-objects-in-image?s_tid=prof_contriblnk

load Image;

pgons=bwlpolyshape(~BW, Visualize=true);

shapes=arrayfun(@classify,pgons(:))

shapes = 8×1 string array
    "Square"
    "Triangle"
    "Rectangle"
    "Rectangle"
    "Circle"
    "Square"
    "Triangle"
    "Square"

function shape=classify(pgon)

    Lengths=vecnorm(diff(pgon.Vertices([end,1:end],:)),2,2); %Side lengths
    
    Lengths(Lengths<max(Lengths)/20)=[]; %Tolerance on shortest side length
    
     N=numel(Lengths); %Number of sides (after tolerance)
     
     if N==3
         
       shape="Triangle";
       
     elseif N==4
 
         shape="Rectangle";
         if max(Lengths)-min(Lengths)<3 %Tolerance on side length equality
             shape="Square";
         end
         
     elseif N>10
         
          shape="Circle";
         
     end
     
     [cx,cy]=centroid(pgon);
     drawpoint(Position=[cx,cy],Label=shape,LabelText="white");
     
end

[Daniel]

im having trouble getting the bwlpolyshape to work, did you just add the downloaded code to the image toolbox codes in here C:\Program Files\MATLAB\R2025b\toolbox\images\images

[Matt J]

It should work if its anywhere on your Matlab path, but I wouldn't recommend adding things to built-in toolbox folders.

Answer 2

[Image Analyst]

Yeah, of course. You can't use things like aspect ratio, area, etc. to classify shapes. Aspect ratio of a rectangle will change as you rotate it. And the shapes, in general, could be of any size and you need to account for that unless you know for a fact that all your test shapes will have the same orientation and size and only the location could be different.

A more robust way to detect simple shapes like yours is to count the number of vertices. See my attached demos.