Steve on Image Processing

April 14th, 2008

The three most common questions I've been hearing about bwlabel, which is used for labeling connected components in a binary image, are about

• Search order
• Relabeling (renumbering the label matrix)
• Correspondence between labels in two different images

Today I'll tackle relabeling.

In my previous post on search order, I showed how to postprocess the labeled objects to modify their order according to various criteria. Now I'll take that a step further and use the sorted output from regionprops to renumber the labels in the label matrix.

Blog reader Trung wanted to know if we could sort lexicographically by centroid, so I'll do that here.

url = 'http://blogs.mathworks.com/images/steve/186/scanned_page.png';
bw = imread(url);
bw = ~bw(1107:1194, 17:135);
imshow(bw, 'InitialMagnification', 'fit')

Here's a false-color view of the label matrix using label2rgb:

L = bwlabel(bw);
rgb = label2rgb(L, 'jet', [.95 .95 .95], 'shuffle');
imshow(rgb, 'InitialMagnification', 'fit')

Now let's use regionprops to get the centroids for each object. To facilitate the relabeling step, I'll get the 'PixelIdxList' for each object.

s = regionprops(L, {'Centroid', 'PixelIdxList'});

Next, sort lexicographically by the centroids, sorting first by the vertical coordinate.

centroids = cat(1, s.Centroid);
[sorted_centroids, sort_order] = sortrows(fliplr(centroids));
s2 = s(sort_order);

To visualize the sorted object order, we can display the object numbers on top of the objects like this:

% First, make an image with a light gray background instead
% of a black background, so that the numbers will be visible
% on top of it.
I = im2uint8(bw);
I(~bw) = 200;
I(bw) = 240;
imshow(I, 'InitialMagnification', 'fit')

% Now plot the number of each sorted object at the corresponding
% centroid:
hold on
for k = 1:numel(s2)
   centroid = s2(k).Centroid;
   text(centroid(1), centroid(2), sprintf('%d', k));
end
hold off

We sorted the output of regionprops, but we haven't touched the label matrix itself. We can relabel it according to the sort order by using linear indexing and the PixelIdxList of each object. An object's PixelIdxList is a vector of linear indices for the pixels belonging to the object. Here's the relabeling loop:

for k = 1:numel(s2)
   kth_object_idx_list = s2(k).PixelIdxList;
   L(kth_object_idx_list) = k;
end

In MATLAB 7.6 (R2008a), the publish feature lets you capture multiple graphics from within a loop. I'll use that new capability to show the location of the first few relabeled objects in L.

for k = 1:5

   imshow(L == k)

end

Soon I'll write another post that shows how how to match up labels in objects that overlap between two images.

Get the MATLAB code

Published with MATLAB® 7.6

09:50 UTC | Posted in Connected components | Permalink |

You can follow any responses to this entry through the RSS 2.0 feed. You can skip to the end and leave a response. Pinging is currently not allowed.

9 Responses to “Relabeling a label matrix”

1. Swetha replied on August 7th, 2008 at 04:08 UTC :
   

   Hello Steve,

   The posts on connected components is very useful.I have tried relabelling the objects with two images an image and i just clipped some portion and tested . I get two different objects ,an object recognised in one image and in another image it is not an object . How is it possible?
   -swetha

2. Swetha replied on August 7th, 2008 at 04:10 UTC :
   

   Hello,

   If you are confused with the two images taken ,i have taken image1- some image ,image2-clipped version of image1.
   -swetha

3. Steve replied on August 13th, 2008 at 11:30 UTC :
   

   Swetha—Your question is pretty vague. Can you be more specific?

4. azhagumani replied on September 16th, 2008 at 01:12 UTC :
   

   i want to merge the segments having the same average intensity value and adjacent to each other

5. Steve replied on September 16th, 2008 at 06:46 UTC :
   

   Azhagumani—The Image Processing Toolbox does not have such a function, so you’ll need to code it up yourself.

6. shahn replied on June 12th, 2009 at 17:11 UTC :
   

   Dear Steve

   I have tried relabelling objects in my binary image, using
   region groups but the labelling is incohereant.

   how can I relabel my binary objects row wise as :

   1 2 3 4 5 6 7
   8 9 10 11 12 13 etc ?????? in my case the objects represent atoms.
   Regards Shahn

7. Steve replied on June 12th, 2009 at 17:13 UTC :
   

   Shahn—See this post.

8. Piyush replied on October 24th, 2009 at 17:59 UTC :
   

   Hi, I have a labeled matrix but one thing I would like to do is create text labels on the image matrix itself without having to open up a figure and plot text labels on them because I am trying to do it in a parallel process, and have it go really fast. For example I have circles that I am checking for their circularity and I just want to overlay text in the image itself over the circle with the circularity value. Is this possible with the current toolset Matlab provides? If not what is the recommended way to implement this?

   Thanks,
   Piyush

9. Steve replied on November 9th, 2009 at 18:22 UTC :
   

   Piyush;mdash;There’s no good way in MATLAB to render text except via graphics commands into a figure window. However, you can do it into an invisible figure and then use the print function to output to an image file. I don’t know whether the performance will be suitable for you or not.

Leave a Reply

<pre class="code">
a = magic(3);
sum(a)
</pre>

These postings are the author's and don't necessarily represent the opinions of The MathWorks.