how do i assign a color to the same type of shape

Now that each shape is classified into its group, how would i go about to add color to each shape, each shape must be colored according to group i.e squares all blue, circles all red,but shape that don't fall into the classification should be black in color. I used RGB2 below but i cant add the shapes together into an image with a white background. Is there any other method???

RGB2 = label2rgb(L==n, 'copper', 'w', 'noshuffle');

eg. squares=blue

eg. circles=r

ed

[Adi Pamungkas]

You can see at https://pemrogramanmatlab.wordpress.com/2016/07/02/pengolahan-citra-digital-untuk-mendeteksi-warna-dan-bentuk-obyek/

Answer 1

[Doug Hull]

If you make this a binary image, you can run it through regionprops. There are a variety of metrics in there that should be usefull in identifying equal shapes. Without much more information, I think this will be enough to get you in the right direction.

[jason]

thank you, but i am not allowed to use region prop command.

Answer 2

[Image Analyst]

Once you've identified the shapes, which we've talked about in your other question, you can use a for loop to reassign your labeled region to the desired number, like 1 for squares = blue, 2 for circles = red, and so on. Then just create the colormap and pass that plus the labeled image in to ind2rgb() to get an RGB image with shapes having the desired color.

[jason]

Where about would i put that code Image Analyst, note that i have to get a black and white binary output with the shape assigned a number to it which we did yesterday, now i have to group the shapes and assign each group a different color. How should i go about doing it? Please note i have a few hours before i present and do not have matlab on this current pc.

[jason]

here is my code with the classified shapes, take a look @ it please, where would i add the color to codes.

Answer 3

[DGM]

It's no surprise OP got lost. That's an awful lot of duplicated code, unused code, and misleading variable names. I got lost.

This is an attempt to clean it up so that it runs at least as much as it did before. There are still a lot of things that could be simplified, fixed, or otherwise improved.

The code as given wouldn't run with the given image, since the edge detection generates a bunch of spurious blobs. That's probably because the input is a crusty JPG.

clearvars                                   % clean up
close all
clc

fontsize = 8;                               % used for figure labels
textbgc = 'y';

RGB = imread('test_shapes.jpg');            %read the loaded image

sharp=imsharpen(RGB);                       %sharpen the read image
grayscale=rgb2gray(sharp);                  %convert from rgb image to grayscale format
h = fspecial('gaussian',40,1);              %create a smoothing filter
filtered = imfilter(grayscale,h);           %filtering
Edgecanny = edge(filtered,'canny');         %edge detection
X = imfill(Edgecanny,'holes');              %fill in with holes

% get rid of all the tiny blobs created by doing edge detection on a JPG
Bw = bwareaopen(X,1500);

% get the label image
[L numblobs] = bwlabel(Bw);

% define shape class names
shapeclasses = {'Circle','Square','Triangle','Rectangle','Pentagon', ...
    'Hexagon','Ellipse','Octagon','Diamond','Unknown'};

% preallocate
blobcentroids = zeros(numblobs,2); % the [x y] centroid location for each blob
blobclasses = zeros(numblobs,1); % the class number for each blob
blobaream = zeros(numblobs,1); % the unrounded area for each blob (in sq m)

for n=1:numblobs
    %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
    % SHAPE FACTORS
    thisblobonly = L==n; % do it once instead of 14 times 
    [row, col] = find(thisblobonly);
    
    % Find Bounding Box
    sx = min(col)-0.5; %starting position for column
    sy = min(row)-0.5; %starting position for row
    breadth = max(col)-min(col)+1; %gives breadth of the box
    length = max(row)-min(row)+1; %gives length of the box
    BBox = [sx sy breadth length]; %Bounding Box value for the range of n
    
    % Find area ratio
    blobarea = bwarea(thisblobonly); %calculates the area of each binary component
    blobaream(n) = round(blobarea) * 0.001; %converting from mm to m
    bboxarea = (length*breadth); %calculates the area of each bounding box
    extent = blobarea/bboxarea; %calculates the ratio of area's
    fprintf('Extent: %.4f\n',extent)
    
    % Find Perimeter
    thisboundary = bwboundaries(thisblobonly); %Find the boundary of the labeled components boundary pixels
    c = cell2mat(thisboundary(1)); %converts the cell to a matrix
    Perimeter = 0; %declaring a start at zero for sum
    for i = 1:size(c,1)-1 %creates a loop that extracts perimeter of each binary component
        Perimeter = Perimeter+sqrt((c(i,1)-c(i+1,1)).^2+(c(i,2)-c(i+1,2)).^2);
    end
    fprintf('Perimeter: %.3f\n',Perimeter); %display perimeter
    
    % Find Roundness
    Roundness = (4*blobarea*pi)/Perimeter.^2; %defines how close the binary component is to a circle
    fprintf('Roundness: %.3f\n',Roundness); %display roundness factor
    
    % Find aspect ratio instead of distance between first three vertices
    aspectratio = min(BBox(3:4))/max(BBox(3:4)); % normalized
    fprintf('Aspect: %.3f\n',aspectratio);
    
    % Find Centroid
    X = mean(col); %calculates average of columns
    Y = mean(row); %calculates average of rows
    blobcentroids(n,:) = [X Y]; %matrix of centroid dimensions
    fprintf('Centroid: [%d %d]\n',round(blobcentroids(n,:))); %display centroid
    
    %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
    % SHAPE CLASSIFICATION
    if extent>=0.782 && extent<=0.800 && Roundness>=0.909
        blobclasses(n) = 1; % CIRCLE
        
    elseif aspectratio>=0.9 && extent>=0.97
        blobclasses(n) = 2; % SQUARE
        
    elseif extent>=0.526 && extent<=0.550 && Roundness>=0.541 && Roundness<=0.559
        blobclasses(n) = 3; % TRIANGLE
        
    elseif aspectratio<0.9 && extent>=0.97
        blobclasses(n) = 4; % RECTANGLE
        
    elseif extent>=0.694 && extent<=0.714
        blobclasses(n) = 5; % PENTAGON
        
    elseif extent>=0.758 && extent<=0.769
        blobclasses(n) = 6; % HEXAGON
        
    elseif extent>=0.782 && extent<=0.800 && Roundness<0.909
        blobclasses(n) = 7; % ELLIPSE
        
    elseif extent>=0.833 && extent<=0.909
        blobclasses(n) = 8; % OCTAGON
        
    elseif extent>=0.526 && extent<=0.535 && Roundness>=0.714 && Roundness<=0.769
        blobclasses(n) = 9; % DIAMOND
        
    else
        blobclasses(n) = 10; % NOT CLASSIFIED SHAPE
        
    end
    
    % dump another message into the console
    fprintf('Blob %d is a %s\n\n',n,lower(shapeclasses{blobclasses(n)}));
end

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% create a bunch of subplots 
% doing the junk with blind axes position manipulation is a nightmare
% it's entirely dependent on the relationship between the aspect ratio of the figure window
% and the aspect ratio of the images.  
% there's no reason to expect that the titles will even be visible.
% the modern suggestion would be to use tiledlayout, but that wasn't an option in 2014.
% FEX subplot_tight() would have probably been sufficient at the time

subplot_tight(2,2,1)
imshow(RGB)
title('Original Image');

subplot_tight(2,2,3)
% idk why the original plotted the input image twice
% i'm just going to plot the binarized image
% since that's why this whole mess wouldn't work in the first place
imshow(Bw)
title('Binarized Image');

subplot_tight(2,2,2)
imshow(Bw)
title('Classified Shapes');
for k = 1:numblobs
    ht = text(blobcentroids(k,1),blobcentroids(k,2),shapeclasses{blobclasses(k)});
    set(ht,'horizontalalignment','center','fontsize',fontsize,'backgroundcolor',textbgc)
end

subplot_tight(2,2,4)
imshow(Bw)
title('Area in m^2');
for k = 1:numblobs
    ht = text(blobcentroids(k,1),blobcentroids(k,2),sprintf('%.3f',blobaream(k)));
    set(ht,'horizontalalignment','center','fontsize',fontsize,'backgroundcolor',textbgc)
end

%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% create an image with blobs colored by class

% this could also have been done in the main loop
labeledbyclass = zeros(size(Bw),'uint16'); % integer-class
for k = 1:numblobs
    thisblobonly = L==k;
    labeledbyclass(thisblobonly) = blobclasses(k);
end

% labeloverlay() was not available in 2014
% treat the new label image as an indexed image
% indexed image mapping depends on the numeric class
% for float inputs, mapping starts at 1
% for integer-class inputs, mapping starts at 0
CT = brewermap(numblobs,'paired'); % create the color table
CT = [0 0 0; CT]; % include a black row for the background
coloredbyclass = ind2rgb(labeledbyclass,CT);

figure
imshow(coloredbyclass)

This uses subplot_tight() and brewermap() from the File Exchange. Both would have been available at the time, so I think that's fair. I used brewermap() here simply as a quick way to get a categorical colormap that's long enough for the 10 classes given. The built-in lines() map is not. There would have been alternatives to both.

Note that the classification doesn't actually work very well. I didn't bother to fix that. This was mostly just an exercise creating the colored label image with period-correct tools, but it got sidetracked by simplifying existing code without actually digging into OP's classification approach. Similarly, I'm not touching the question of whether doing edge detection is even an appropriate part of the binarization task. Maybe I'll come back to this some other day.

EDIT: Nevermind; I went ahead and made the classification work at least. It's still not great, but at least it works for this image, and the console dump is helpfully formatted.

You might ask why there aren't any circles or squares detected. That's because there aren't any. Test it with a different image: