Detecting and counting the number of scales (circular or psuedo circular objects in the image) and drawing boundaries around the scales or circular objects.

I have asked this question previously one the forum and got a few answers for the same, however they do not really answer the core aspects of what I want to solve. So, I am asking a modified version of the same question again.

I have multiple grayscale images, which have circular or psuedo circular objects within them. I need to detect them, draw boundaries around them and then count the number of these objects in each image.

I am attaching one grayscale image for your reference.

Any inputs will be highly appreciated.

This is the code that I had tried out.

Guys, can anyone give any suggestions, I would really appreciate some help as I am on a ticking clock.

I have attached the photos with objects marked in red.

X = imread("f_000047.tif");
% Adjust data to span data range.
X = imadjust(X);
X  = imgaussfilt(X,1.4);
% Threshold image - manual threshold
BW = X > 164;
% Create masked image.
fontSize = 8;
maskedImage = X;
maskedImage(~BW) = 0;
SE = strel("disk",7); % Generating a disk-type structuring element of radius 10 for dilation
maskedImage = imdilate(maskedImage,SE); %Filling up holes or inaccuracies in the intial BGMask
%maskedImage = imrotate(maskedImage,-90);
subplot(3,1,1); 
imshow(maskedImage);
title("MaskedImage","FontSize",fontSize);
%----------------------------------------------------------------------------------------------------------
%Finding out higher SD regions across the background
ImgSD = stdfilt(X,ones(13)); %Standard deviation (SD) filtering based on varying SDs across the image 
ImgSD(~maskedImage)=0; %Removing the background
subplot(3,2,2); 
imshow(ImgSD,[]);
subplot(3,2,3);
histogram(ImgSD,100);
[x,y] = size(ImgSD);     
xGrad = zeros(x,y);%Horizontal gradient matrix definition
yGrad = zeros(x,y);%Vertical gradient matrix definition
Tan_ver_add = size(x,y);
Tan_horz_add = size(x,y);
Tan_ver_sub = size(x,y);
Tan_horz_sub = size(x,y);
for i = 1:x-1
     for j = 2:y-1
        xGrad(i,j) = (ImgSD(i,j+1)-ImgSD(i,j-1))/2; % finding the horizontal gradient across the image (Method 1)
     end
end
h_grad = (circshift(ImgSD,[0,1])-circshift(ImgSD,[0,-1]))/2; %Method 2
for j = 1:y-1
    for i = 2:x-1
        yGrad(i,j) = (ImgSD(i+1,j)-ImgSD(i-1,j))/2; % finding the vertical gradient across the image (Method 1) 
    end
end
v_grad = (circshift(ImgSD,[1,0])-circshift(ImgSD,[-1,0]))/2; %Method 2
%-----------------------------------------------------------------------------------------------
%Gradient Modulus
grad_tot_mod =sqrt((h_grad).^2+(v_grad).^2); % finding the magnitude of gradient across the image (Method 2)    
GradMod = sqrt((xGrad).^2+(yGrad).^2); %Method 2 for Gradient Modulus
%---------------------------------------------------------------------------------------------
%Calculating double derivative in the horizontal and vertical directions as
%a second parameter for reducing the area of interest
double_der_h = zeros(x,y); %Horizontal double derivative matrix definition
double_der_v = zeros(x,y); %Vertical double derivative matrix definition
for i = 1:x-1
    for j = 2:y-1
        double_der_h(i,j) = (ImgSD(i,j+1)-2*ImgSD(i,j)+ImgSD(i,j-1)); % finding double derivative across x-direction in the image
    end
end
for j = 1:y-1
    for i = 2:x-1
        double_der_v(i,j) = (ImgSD(i+1,j)-2*ImgSD(i,j)+ImgSD(i-1,j)); % finding double derivative across y-direction in the image
    end
end
double_der_tot = double_der_h+double_der_v;% Can we do scalar addition of the individual components of the double derivative or do we need to calculate the modulus
%Should I use a Laplacian operator instead?
%doubleDMod = sqrt((double_der_h).^2+(double_der_v).^2);
subplot(3,3,1);
imshow(X);
hold on
for i = 1:x
    for j = 1:y
            if (grad_tot_mod(i,j)>0.1 && grad_tot_mod(i,j)<10) %imposing the gradient condition to plot markers
                if (double_der_tot(i,j)~=0) % imposing double derivative condition to plot markers %Ask about a different condition >=0
                    if ImgSD(i,j)>=0
                        Tan_horz_add= atan2d(h_grad(i+5,j),v_grad(i+5,j));
                        Tan_horz_sub= atan2d(h_grad(i-5,j),v_grad(i-5,j));
                        Tan_ver_add= atan2d(h_grad(i,j+5),v_grad(i,j+5));
                        Tan_ver_sub= atan2d(h_grad(i,j-5),v_grad(i,j-5));
                        %Changed the angle definition from
                        %(v_grad(),h_grad()) to (h_grad(),v_grad())
                        if Tan_horz_add<0
                            Tan_horz_add= Tan_horz_add+180;
                        end
                        if Tan_ver_add < 0
                            
                  Tan_ver_add = Tan_ver_add+180;
                        end
                        if Tan_horz_sub < 0
                            Tan_horz_sub = Tan_horz_sub+180;
                        end
                        if Tan_ver_sub < 0
                            Tan_ver_sub = Tan_ver_sub+180;
                        end
                       if ((abs(Tan_ver_add-Tan_ver_sub)>= 5 && abs(Tan_ver_add-Tan_ver_sub)<15)||(abs(Tan_ver_add-Tan_ver_sub)>=95 && abs(Tan_ver_add-Tan_ver_sub)<120)) 
                           if ((abs(Tan_horz_add-Tan_horz_sub)>=5 && abs(Tan_horz_add-Tan_horz_sub)<15))
                                plot(i,j,"*")
                           end
                       end
                    end
                end
            end
    end
end
hold off

[Image Analyst]

Edit the post to include an image with the regions you want to find outliend in red.

[Image Analyst]

Some of those regions are not well closed. If my previous answer did not work then I suggest you perhaps try SegNet deep learning, or just hand trace the things, because I can't really tell where the things start and stop once you zoom in to a pixel basis. They just don't have clearly defined boundaries. You, or SegNet, can look at intensities plus take into account the local shape of the thing and try to give you roughly elliptical things even if they don't have a clearly defined boundary.

[Atreya Danturthi]

Thank you for your suggestion.

I thought of using Deep Learning, but the issue is I do not really hold a lot of knowledge in the field and moreover I have limited amount of time, so I am not sure as to if I will be able to learn the required aspects of Deep Learning and apply the same successfuly.

On the otherhand, I have roughly 300 images with me, therefore manually looking into each of the images is not really feasible.

If you do have knowledge on deep learning, could you tell as to what aspects do I need to learn to give it shot?

[Image Analyst]

With 300 images you could hand trace all those images in about 5 hours at a minute per image. It would take you longer than that to get a deep learning solution and you don't have time for that. Use drawfreehand() and you could get it all done in an afternoon.

[Atreya Danturthi]

Oh, I see. Thank you for the suggestion, I will surely try it out.

If I hand trace the boundaries and they are not precise, wouldn't it affect my analysis?

I need to get the areas, diameter and a few more parameters of those objects.

[Atreya Danturthi]

I have tried to use the freehand method, but it shows an error stating that your image is not rgb type and therefore the required operation cannot be performed.

When I converted the image to rgb, it says "Number of elements must not change. Use [] as one of the size inputs to automatically calculate the appropriate size for that dimension".

When I applied [], it says too many inputs.

[Image Analyst]

It works fine with gray scale images. Here's proof:

grayImage = imread('cameraman.tif');
imshow(grayImage, []);
drawnow;
g = gcf;
g.WindowState = 'maximized'
uiwait(msgbox('Draw a freehand curve'));
h = drawfreehand
% Rplace by a curve.
xy = h.Position;
x = xy(:, 1);
y = xy(:, 2);
% Tack first point onto last so the curve will be closed.
x = [x; x(1)];
y = [y; y(1)];
hold on;
delete(h);
plot(x, y, 'y-', 'LineWidth', 3);

[Atreya Danturthi]

"Error using images.internal.imageDisplayValidateParams>validateCData (line 118)

Multi-plane image inputs must be RGB images of size MxNx3.

Error in images.internal.imageDisplayValidateParams (line 30)

common_args.CData = validateCData(common_args.CData,image_type);

Error in images.internal.imageDisplayParseInputs (line 79)

common_args = images.internal.imageDisplayValidateParams(common_args);

Error in imshow (line 253)

images.internal.imageDisplayParseInputs({'Parent','Border','Reduce'},preparsed_varargin{:});

Error in Freehandmethod (line 2)

imshow(grayImage, []);

This is the error that showed up when I replaced the cameraman image with one of my images.