Better skeletonization of an image to found branchpoints

Hi,

I have an image of an grid and want to find the 12 branchpoints of it. But the skeleton is really "rough" so I have to many branchpoints. I tried different "clean up" methods but I can't clean it up enough. Can someone help me with my problem, please.

Grid = imread('grid.jpg');
gray_grid = rgb2gray(Grid);
binary_grid = im2bw(gray_grid);
inverse_grid = imcomplement(binary_grid);
grid_skel = bwmorph(inverse_grid,'skel',inf);
%%clean skel
skel_1 = bwmorph(grid_skel,'spur',inf);
skel_2 = bwmorph(skel_1,'clean',inf);
%%find branchpoints
branchpoints = bwmorph(skel_2,'branchpoints');
[rowsBP, columnsBP] = find(branchpoints)
%%show branchpoints
hold on;
for k = 1 : length(rowsBP);
  plot(columnsBP(k), rowsBP(k), 'r+', 'MarkerSize', 8, 'LineWidth', 1);
end

Answer 1

[Massimo Zanetti]

Dilate your image before skeletonize it. Try this:

%read image
G = imread('grid.jpg');
G = im2bw(rgb2gray(G));
%dilate image complement
I = imdilate(imcomplement(G),strel('disk',15));
figure; imshow(I)
%skeletonize, and remove spur branches
S = bwmorph(I,'skel',inf);
S2 = bwmorph(S,'spur',inf);
figure; imshow(S2);
%find branchpoints, dilate them to see them in image
B = bwmorph(S2,'branchpoints');
B2 = imdilate(B,strel('disk',10));
figure; imshow(B2);

[tsaiwu]

thanks for your help thats a great way and I get a much finer skeleton, but after this I want to know where the single branchpoints are (x,y). And I have still to many branchpoints, only in the image it appears as one branchpoint. My goal is to finde the single branchpoints (x,y coordinates) in the grid.

Ok, found a way to do it. I do it with circle detection and find the center points of it. Now I will try if this code works with a other painted grid.

Edit: Code doesn't work so well for the second grid.

clc; clear all;
%read image
G = imread('grid.jpg');
G = im2bw(rgb2gray(G));
%dilate image complement to thick it up
I = imdilate(imcomplement(G),strel('disk',13));
%skeletonize, and remove spur branches
S = bwmorph(I,'skel',inf);
S2 = bwmorph(S,'spur',inf);
S3 = bwmorph(S2,'clean',inf);
%find branchpoints, dilate them to see them in image
B = bwmorph(S3,'branchpoints');
B2 = imdilate(B,strel('disk',10));
[centers, radii] = imfindcircles(B2,[10 30],'ObjectPolarity','bright');
figure; imshow(G);
hold on
plot(centers(:,1),centers(:,2), 'r+', 'MarkerSize', 10, 'LineWidth', 2);
centers

[Image Analyst]

Why are you dilating and calling infindcircles()? Simply call find() and plot

[y, x] = find(B);
plot(x, y, 'r+', 'MarkerSize', 10, 'LineWidth', 2);

[tsaiwu]

because bwmorph(branchpoints) finds to many branchpoints, I want the coordinates of the intersections points of the grid.

[Massimo Zanetti]

I have dilated the branchpoints (image B2) only to enhance their visualization.. In matrix B you have only the right pixels on.. to get the coordinates (i,j index) just call [I,J]=find(B).

There is no need to run additional code after the piece of code I wrote.

[tsaiwu]

If i ran find[B] it gives me to many branchpoints backs, like I said before. I want to find only the intersections point of the grid, for example in the first grid 12 points.

And do you have an idea to get the intersection points of the second grid? thanks

[Massimo Zanetti]

Ok, I got it. So, capture the centroids of the cirlces using regionprops. Like this:

I = imdilate(imcomplement(G),strel('disk',15));
S = bwmorph(I,'skel',inf);
S2 = bwmorph(S,'spur',inf);
B = bwmorph(S2,'branchpoints');
B2 = imdilate(B,strel('disk',10));
s = regionprops(B2,'centroid');
centroids = cat(1, s.Centroid);
figure;
imshow(G); hold on;
plot(centroids(:,1),centroids(:,2),'r*');
hold off;