Why doesn't bwconncomp detect each connected component as expected?

Hi everyone!

I have the following binary image M

I have a problem with bwconncomp. My aim is to generate Voronoi diagrams using as input not single points but blobs/connected components. However the command bwconncomp fails to detect each connected component as expected; as you can see from the following figure, it seems to calculate the centroid of every cluster of pixels, which is not what I want because I would obtain the voronoi diagram around the yellow shapes of the first figure, not around a single centroid point. How can I solve this? I attach an extract of my code and a zip file containing the functions and the main code called codice_gvd.m. Thanks in advance

%% Pose

% Initial Pose
xs = 70; ys = 76; % valori iniziali 55 e 33
% Goal Pose
xg = 50; yg = 50; % valori iniziali 50 e 50
% Pose Matrix
pose = zeros(sz(1),sz(2));
pose(ys,xs) = 1;
pose(yg,xg) = 1;


%% Voronoi Road Map

figure('color','k');
hold on;
set(gcf, 'InvertHardCopy', 'off');

CC = bwconncomp(M); 
p = CC.PixelIdxList; 
m = zeros(length(p), 2); 
  
for ii = 1:length(p) 
  
      [r, c] = ind2sub(size(M), p{ii}); 
  
      m(ii,:) = [mean(c), mean(r)]; 
  
  end
  
  voronoi(m(:,1), m(:,2));
  title({'Voronoi Road Map';'* - Represents Starting and Goal Position '},'color','w');
  spy(pose,'*r')

[Walter Roberson]

CC = bwconncomp(hit);

we need hit for testing purposes. Also if hit is derived from something else it would help to see the logic chain

[Loren99]

@Walter Roberson I have uploaded the zip file containing the main code 'codice_gvd' and other folders that are the functions and class functions needed to run the main code

Answer 1

[Image Analyst]

Your binary image, hit, is a 1-D vector, not really an image. So not sure why you seem to be expecting that it's a 2-D image.

[Loren99]

@Image Analyst right! In fact I have corrected and now I have used the binary image M as input, not hit. However now the problem is that bwconncomp seems to calculate the centroid of every group of pixels, not the connected component in order to form a sort of polygon

[Image Analyst]

What do you want? So let's say that in general you have a binary image with several irregularly shaped blobs in it. You can get their boundaries, centroids, areas, pixel locations within the blob, etc. In other words, you can measure a whole bunch of things about those blobs. What would you like to measure? The skiz?

https://smil.cmm.minesparis.psl.eu/publis/Cours-Serra/chap-09-en.pdf

[Loren99]

@Image Analyst exactly! I would like to obtain the skiz (voronoi diagram) around these several irregularly shaped blobs. I have tried several algorithms: with the algorithm used here https://it.mathworks.com/matlabcentral/answers/1734135-how-to-eliminate-some-useless-lines-in-an-image there is the problem that even if I obtain the lines around the irregular shapes, I do not know how to do a pruning to remove the useless lines (the ones that are directly attached to the shapes). So I have tried another algorithm (the one that I have posted here and in the zip file) using bwconncomp, thinking that in this way creating connected components, maybe the in-built function voronoi does not generate that useless lines. Instead what happens is that bwconncomp generates the centroids of the irregular shapes and the voronoi diagram is created around these points, not around the shapes. I hope this is clear now, sorry

[Image Analyst]

See Steve's blog where he computes the skiz:

https://www.mathworks.com/help/images/marker-controlled-watershed-segmentation.html