I want to connect the ends of the white contour in the image to form a complete circle, eliminating the gaps.

Hi!

In this image, I want to connect the ends of the white contour to form a complete circle. Currently, there is a break in its structure, and my goal is to detect the endpoints and link them to obtain a closed shape. I want to use the bwskel function in MATLAB.

img = imread('imag5_Seg.png');
bw = imbinarize(img); 
bw = logical(bw);
schelet = bwskel(bw);
figure();
imshow(schelet);

Answer 1

[Image Analyst]

Here's another way:

%-----------------------------------------------------------------
% Read in and prepare the sample image.
binaryImage = imread('ring gap.png');
[rows, columns, numberOfColorChannels] = size(binaryImage)

rows = 1036

columns = 1092

numberOfColorChannels = 3

% It's RGB so convert to binary.
if numberOfColorChannels > 1
    binaryImage = binaryImage(:, :, 1) > 0;
end
% The first 11 lines in the supplied sample image are not zero for some reason.  Erase them.
binaryImage(1:11, :) = false;
% Save the original binary image.
originalBinaryImage = binaryImage;
% Erase bottom half to give us two blobs
binaryImage(round(columns/2) : end, :) = false;
% Make sure there are only two blobs.  Take the biggest 2.
binaryImage = bwareafilt(binaryImage, 2);
% Label the image so that each blob has a unique ID %
[labeledImage, numBlobs] = bwlabel(binaryImage);
imshow(labeledImage, []);
axis('on', 'image');

%-----------------------------------------------------------------
% EDGE LINKING CODE
% Find the rows and columns of blob #1
[rows1, cols1] = find(ismember(labeledImage, 1));
% Find the rows and columns of blob #2
[rows2, cols2] = find(ismember(labeledImage, 2));
% Find the distance between all pixels in blob1 and all pixels in blob2
distances = pdist2([cols1, rows1], [cols2, rows2]);
% Find the min distance
minDistance = min(distances, [], 'all')

minDistance = 88.6848

% Find the index where the min distance occurs
[mRow, mCol] = find(distances == minDistance)

mRow = 5258

mCol = 32

% Get the coordinates of where that is.
x1 = cols1(mRow)

x1 = 265

y1 = rows1(mRow)

y1 = 370

hold on;
% plot(x1, y1, 'm.', 'MarkerSize', 30) % Plot magenta spot at the first point.
x2 = cols2(mCol)

x2 = 309

y2 = rows2(mCol)

y2 = 293

% plot(x2, y2, 'c.', 'MarkerSize', 30) % Plot cyan spot at the second point.
% Draw a line in the overlay between the endpoints.
plot([x1, x2], [y1, y2], 'r-', 'LineWidth', 3)

●

%-----------------------------------------------------------------
% BURN LINE INTO ORIGINAL IMAGE.
% If you have the Computer Vision Toolbox, use insertLine.
% If you don't, use linspace to get the 9-connected coordinates and then a loop to burn in the pixels.
numPixels = round(minDistance * 1.5); % Make sure we have enough sample points to not have any gaps.
x = round(linspace(x1, x2, numPixels));
y = round(linspace(y1, y2, numPixels));
% Loop over them, setting pixels in the original image.
for k = 1 : numPixels
    originalBinaryImage(y(k), x(k)) = true;
end
% Display the final image.
figure;
imshow(originalBinaryImage);
title('Final Edge L:inked Image', 'FontSize', 20);

Sorry it doesn't look connected here in the web browser, but trust me, it really is. If you do it in MATLAB and then zoom in, you'll see the line connecting the gap is a continuous 8-connected line of pixels.

If you want the line a lot thicker, then write the line to a totally blank image, use imdilate to thicken it as much as you want, and then OR that back in to your original image.

[FLOREA]

Thank you very much, I think this method will be helpful for me. I will tell you my entire problem from the beginning. So, I started with a CT image (attached below) and I am trying to segment the affected region to identify a possible lesion. To do this, I want to perform segmentation within the bright area (the skull bone). The problem is that due to missing pixels, the contour of this region is not completely closed. This creates difficulties in applying segmentation techniques because the region is not perceived as a continuous shape.

[Image Analyst]

I'm not sure why the skull region must be closed. Why should whether it's closed or not affect your segmentation of the brain matter and the bright or dark region at the center of the brain?

Answer 2

[Mathieu NOE]

hello

sorry , this is a "no image processing" code - it may or may not correspond to your needs , but it does the job !

hope it helps

you will have to download this Fex submission for smoothing (I like it very much) but yiu ca of course use another smoother

smoothn - File Exchange - MATLAB Central

result :

code :

%% load file
file_name='image.png';
img=imread(file_name);
if size(img,3)==3
    img=rgb2gray(img);
end
% binarize image
outpic = zeros(size(img));
ind = find(img>128);
outpic(ind) = 1;
[y,x] = ind2sub(size(img),ind); % get x,y points coordinates of the white area 

%% method 1
centroid_x = mean(x);
centroid_y = mean(y);
[theta,r] = cart2pol(x-centroid_x,y-centroid_y);
% sort theta in ascending order
[theta,ind] = sort(theta);
r = r(ind);
% remove duplicates 
[theta,IA,IC] = unique(theta);
r = r(IA);

% take the average of r within an angle of range dt (if no data then NaN output)
Npoints = 100;
theta_new = linspace(min(theta),max(theta),Npoints);
dt = mean(diff(theta_new));

for k = 1:Npoints
    ind = (theta>=theta_new(k)-dt/2) & (theta<theta_new(k)+dt/2);
    if isempty(ind)
        r_new(k) = NaN;
    else
        r_new(k) = mean(r(ind));
    end
end

% replace / interpolate the NaN values
r_new = fillmissing(r_new,'spline');

% convert to cartesian
[xn,yn] = pol2cart(theta_new,r_new);
% add back centroid info
xn = xn + centroid_x;
yn = yn + centroid_y;

% smoothing (if needed)
Y = {xn,yn};
[zz,s,exitflag] = smoothn(Y,10); 
xn = zz{1};
yn = zz{2};

% closing the curve
xn(end+1) = xn(1);
yn(end+1) = yn(1);


%% final plot
figure(1),
imshow(outpic)
hold on
plot(xn,yn,'r','linewidth',2);
hold off

[FLOREA]

It's not what I'm looking for, but thank you very much for your effort!

[Image Analyst]

Why not? Is it because you insist on using bwskel? If so, then why do you have to use that function?