I have used following code to segment a fundus image but the results are not very clear. The test4.jpeg fill is the original fundus images that I have used .please help

I = imread(test4.jpeg);

Unable to resolve the name 'test4.jpeg'.

L = I(:,:,2);
ig = adapthisteq(L,'Numtiles',[100 10],'ClipLimit',0.005);
ig1 = imcomplement(ig);
se = strel('ball',100,100);
godisk = ig1 - imopen(ig1,se);
Si = size(ig);
ig = godisk*10;
ig = double(ig);
ig(ig>255)=255;
orig(:,:,:) = ig;
result = imbinarize(fmd(orig));
x = bwareaopen(result,3000);
result2 = bwmorph(x,'spur',inf);
figure
imshowpair(result2,I,'montage');
imwrite(result2,'fmd/P1_LF.tif');
%%%% following is the output result of this code

% 

%%%%%%%%%%  FMD function used in the above code %%%%%%%%%%%%%
function result = fmd(I)
%imshow(I);
% Settings
plot_results = true;
maximalized_plot = true;
count = 4;
%% Preparing input signals
s = size(I);
m1 = mod(s(1),8);
m2 = mod(s(2),8);
q = 0;
if s(1)>s(2)
    if m1 == 0
        ca = 1;
        p = ceil(s(1)-s(2));
        X = zeros(s(1),s(1));
        X(:,p+1:p+s(2)) = I(:,:);
    else
        rem = 1;
        ss = s(1);
        while(rem ~= 0)
            ss = ss + 1;
            rem = mod(ss,8);
        end
        ca = 1;
        p = ceil(ss-s(2));
        q = ceil(ss-s(1));
        X = zeros(ss,ss);
        X(q+1:q+s(1),p+1:p+s(2)) = I(:,:);
    end
else
    if m2 == 0
        ca = 2;
        p = ceil(s(2)-s(1));
        X = zeros(s(2),s(2));
        X(p+1:p+s(1),:) = I(:,:);
    else
        rem = 1;
        ss = s(2);
        while(rem ~= 0)
            ss = ss + 1;
            disp(ss);
            rem = mod(ss,8);
        end
        ca = 2;
        p = ceil(ss-s(1));
        q = ceil(ss-s(2));
        X = zeros(ss,ss);
        X(p+1:p+s(1),q+1:q+s(2)) = I(:,:);
    end
end
in2D = im2double(uint8(X));
% imshow(in2D);
% figure;
%% DMWT2D and IDMWT2D Example

[out2D, winsize] = DMWT2D(in2D, 'DB2');

% if plot_results
%     f = figure(4);
%     if maximalized_plot
%         f.WindowState = 'maximized';
%     end
%     t = tiledlayout(2, 1);
%     title(t, "DMWT2D and IDMWT2D Example");
%     nexttile;
%     imshow(in2D);
%     title("Original image");
% 
%     nexttile;
%     imshow(abs(out2D));
%     title("Transformed image");
% end


for i = 1:count
    coeff(:,:,i) = out2D((i-1)*winsize+1:i*winsize,:);   
end
for i = 1:count
    for j = 1:count
        subims(:,:,i,j) = coeff(:,(j-1)*winsize+1:j*winsize,i);
    end
end

for i = 1:count
    for j = 1:count
        vecs(:,i,j) = reshape(subims(:,:,i,j).',1,[]);
    end
end
Fs=100e6; 
t = 0:(1/Fs):1-(1/Fs);
%b = zeros(4096,30,4,4);
for i = 1:count
    for j = 1:count
        [CutFrq1] = sp_SetCutOffFreq(vecs(:,i,j),Fs,'ub',10); %uniform filter bank 10
        ImfByFilter = sp_DFTOrthogonalOrFIR_IIR_LINOEP(vecs(:,i,j),t,Fs,CutFrq1,'dct');
        for k=1:length(CutFrq1)-1
            b(:,k)=ImfByFilter{k}; 
        end 
%         [a,b,EnergyLekage]=FMD_Low2High_High2LowSacnning(vecs(:,i,j),Fs,t);
        c = sum(b,2);
        d(:,:,i,j) = reshape(c,[winsize,winsize]); 
    end
end
d = permute(d,[2 1 3 4]);
%d = abs(d);
for i = 1:count
    row1(:,:,i) = d(:,:,1,i); 
    row2(:,:,i) = d(:,:,2,i);
    row3(:,:,i) = d(:,:,3,i);
    row4(:,:,i) = d(:,:,4,i);

end

col1 = [row1(:,:,1) row1(:,:,2) row1(:,:,3) row1(:,:,4)];
col2 = [row2(:,:,1) row2(:,:,2) row2(:,:,3) row2(:,:,4)];
col3 = [row3(:,:,1) row3(:,:,2) row3(:,:,3) row3(:,:,4)];
col4 = [row4(:,:,1) row4(:,:,2) row4(:,:,3) row4(:,:,4)];
img =[col1;col2;col3;col4];
out = IDMWT2D(img, 'DB2');
% figure;
% imshow(imbinarize(out));
if ca == 1
    out = out(q+1:q+s(1),p+1:p+s(2));
else 
    out = out(p+1:p+s(1),q+1:q+s(2));
end
result = out;
end
%%%%%%the test4.jpeg file used is attached below

[KALYAN ACHARJYA]

I = imread('test4.jpeg');

More Let us know if you faced any specific error? The segmention code may work on other test images, ckeck in larger perspectives and do the accuracy analysis

Can you attach DMWT2D file or other custom function files?

[Kanika]

I have attached DMWT2D.I am not getting any other error.The only problem is segmented image is not clear.

[KALYAN ACHARJYA]

DMWT?

Why do we have to ask for files one by one? Please help us by providing all necessary files and data for your assistance? How can any one test code without required data/files?

Which image data base you are referring?

Data base Details (Public data Base)

[Kanika]

I have attached some of the files.This website is not allowing me to attached all the file today .It is telling me to wait for 24 hours .will upload all the files by tommorrow. I am using my own data which I have captured in lab.I have also attached a sample image(test4.jpeg). Thanks for putting afforts.

[Kanika]

All the useful files are in used.zip .I have attached it

Answer 1

[Vijeta]

Hello Kanika,

It seems like the output segmentation result is not very clear and might require some further processing. Here are a few suggestions that might help improve the segmentation result:

1. Try adjusting the parameters of the adapthisteq function to obtain better contrast enhancement. For example, you could try increasing or decreasing the ClipLimit parameter or changing the size of the tiles using the NumTiles parameter.
2. Instead of using a fixed structuring element for morphological operations, try using adaptive structuring elements that can better adapt to the local image features. MATLAB's strel function allows you to create structuring elements of various shapes, sizes, and orientations. For example, you could use a disk-shaped structuring element with a radius that is proportional to the size of the blood vessels in the image.
3. Consider using a combination of morphological operations and other image processing techniques such as edge detection or thresholding to enhance the blood vessels and other features of interest in the image.
4. Try using other segmentation algorithms that are specifically designed for fundus images, such as the Frangi filter, which is a vessel enhancement filter that can be followed by a thresholding step to segment the blood vessels.

You can refer to the following documentation https://www.mathworks.com/matlabcentral/fileexchange/24409-hessian-based-frangi-vesselness-filter

Thanks.

Answer 2

[Image Analyst]

Try

https://www.mathworks.com/matlabcentral/fileexchange/49172-trainable-cosfire-filters-for-curvilinear-structure-delineation-in-images

https://www.mathworks.com/matlabcentral/fileexchange/24409-hessian-based-frangi-vesselness-filter