Segmenting boxes in an image

I'm here to get an advice from all on how to segment boxes in an image.

I have tried using threshold, contour, and watershed algorithm but the result only segments the whole image.

Here is the image that I want to segment. The first box represents a symbol, second box represents a digit, and the third box represents a letter. Why do I need segmentation? I need this segmentation in order to split the boxes according to their representative.

Can anyone advise me?

This is my code

img_o = "/MATLAB Drive/2.png";

% Image loading
img = imread(img_o);

% Create a figure with subplots
figure('units', 'normalized', 'outerposition', [0 0 1 1]);

% Show original image
subplot(2, 3, 1);
imshow(img);
axis off;
title('Original Image');

% Image grayscale conversion
gray = rgb2gray(img);

% Adaptive thresholding
threshold = adaptthresh(gray, 'NeighborhoodSize', 11, 'Statistic', 'Gaussian', 'ForegroundPolarity', 'dark');
binary_img = imbinarize(gray, threshold);

% Show thresholded image
subplot(2, 3, 2);
imshow(binary_img);
axis off;
title('Thresholded Image');

% Perform morphological operations to remove noise
se = strel('square', 3);
opening = imopen(binary_img, se);

% Show opened image
subplot(2, 3, 3);
imshow(opening);
axis off;
title('Opened Image');

% Perform distance transform
dist_transform = bwdist(~opening);
sure_fg = imregionalmax(dist_transform);

% Show sure foreground image
subplot(2, 3, 4);
imshow(sure_fg);
axis off;
title('Sure Foreground Image');

% Perform watershed algorithm
markers = watershed(-dist_transform);
img_copy = img;
img_copy(markers == 0) = 255;

% Show the segmented image
subplot(2, 3, 5);
imshow(img_copy);
axis off;
title('Segmented Image');

% Print the number of segmented objects
num_objects = max(markers(:)) - 1;
subplot(2, 3, 6);
text(0.5, 0.5, ['Number of Segmented Objects: ', num2str(num_objects)], 'HorizontalAlignment', 'center', 'VerticalAlignment', 'middle', 'FontSize', 12);
axis off;

sgtitle('Image Segmentation');

Answer 1

[Angelo Yeo]

clear; close all; clc;
I = imread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/1414384/image.png');
Ibw = ~im2bw(I,graythresh(I));

% getting rid of contents 
comp_obj = bwconncomp(Ibw, 4);

for i = 2:comp_obj.NumObjects
    Ibw(comp_obj.PixelIdxList{i}) = 0;
end

% get the squares inside the rects
comp_obj = bwconncomp(~Ibw, 4);

figure;
imshow(I)
hold on;

[r,c]= ind2sub(size(Ibw), comp_obj.PixelIdxList{2});
scatter(c, r,5,'red','filled','s','MarkerFaceAlpha',0.1)

[r,c]= ind2sub(size(Ibw), comp_obj.PixelIdxList{3});
scatter(c, r,5,'blue','filled','s','MarkerFaceAlpha',0.1)

[r,c]= ind2sub(size(Ibw), comp_obj.PixelIdxList{4});
scatter(c, r,5,'yellow','filled','s','MarkerFaceAlpha',0.1)

[Muhammad Syukri]

Thank you for your help.

But can I know the exact name of your technique here?

What I know, you implement thresholding, then you used 'bwconncomp' function, and used pixel manipulationn with 'Ibw'.

Can you explain more to me?

[Angelo Yeo]

It's nothing but applying morphology 😛 `bwconncomp` detects connected blobs and sorts them by their size.

Answer 2

[Image Analyst]

You can get all three square masks doing this:

mask = grayImage > 128; % Or whatever works.
% Get rid of white surround touching the border.
mask = imclearborder(mask);
% Fill holes.
mask = imfill(mask, 'holes');
% Take the 3 largest blobs.
mask = bwareafilt(mask, 3);

Then to get a mask for each square individually you can use bwlabel and ismember

%--------------------------------------------------------------------------------------------------------
% Get the 3 individual masks and display them.
labeledImage = bwlabel(mask); % Give an ID label to each square.
for k = 1 : 3
    % Get mask 1
    separateMasks{k} = ismember(labeledImage, k);
    subplot(2, 3, 3+k);
    imshow(separateMasks{k});
    impixelinfo;
    axis('on', 'image');
    caption = sprintf('Mask #%d', k);
    title(caption, 'FontSize', fontSize, 'Interpreter', 'None');
end

Full code is in the attached m-file,

[Muhammad Syukri]

Thank you very much. This help too. I have tried and do some modification.