bwlabel changing connectivity of image

I am trying to use BWLabel to label 1 pixel width line segments. These lines are not straight so I am using a connectivity of 8. When I run bwlabel and view the image of the newly labeled skeleton, it actually is changing layout, size and connectivity of the line segments. The input image is a double. I am using Matlab R2009b.

Before BWLabel: http://i.imgur.com/t3Q1C.jpg

After BWLabel: http://i.imgur.com/Sju5B.jpg

Am I asking bwlabel to be too sensitive? I am using this as part of a larger code that is trying to find the angle of these segments by using the segment endpoints, so maintaining separation is important.

Below is the code I used to generate the two supplied images

% A = image to be labeled.
figure(1), imshow(A)
A2 = bwlabel(A,8);
figure(2), imshow(A2)

Answer 1

[Image Analyst]

Your A image is not what you think it is. I'd bet that it is not a logical image, but a gray scale image with most pixels at 255 but some pixels at a much darker value. Do this

imhist(A);

and see what it shows. Do you have only 2 bins at 0 and 1? Or do you have bins at 0, 255, and, say, 20 or some small number?

You can also try this

A = (A == 255);  % Convert to logical image with 1 only where A = 255.

and then label and see what you get.

[Ryan]

I will have to double check tomorrow to be sure, but the image should be logical. It's the result of a skeletonization process using bwmorph commands. I do perform a subtraction (I subtract dilated 'branchpoints' from the 'skeleton' to create the individual line segments), but I am assuming since both were performed with bwmorph that they are still binary images.

[Image Analyst]

Aha! That's your problem. The difference of two logicals is a double. Run this code a few times:

% Make two logical arrays with true in random locations.

m1 = logical(randi(2, [6 6])-1)

m2 = logical(randi(2, [6 6])-1)

diffm = m1 - m2 % This is a double

class(diffm)

% Display it

imshow(diffm);

% Since diffm is a double, imshow will assume diffm

% goes between 0 and 0. The 1's will display as 255.

% Anything from -1 to 0 will display as 0.

% Now label.

bwlabel(diffm)

% Any non-zero things will label including -1.

What is happening is that you are getting a difference array with values -1, 0, and 1. The -1 and 1 will get labeled and show up after labeling. However the -1 won't show up when called with imshow.

To check, call imshow with [] like this:

imshow(A, []);

This will show -1 as 0, 0 as 128, and +1 as 255 since it will scale the whole range of -1 to +1 to the range of 0-255. Once you understand that, all will be clear for you.

[Ryan]

Ahhhh, thank you very much. You and Walter Roberson have taught me a lot. I have not had any training in image processing and am learning this on the fly, mostly by reading posts on here. I picked Matlab over other languages because of the image processing toolbox and the user community. Good stuff.

[Ryan]

I also utilize 'help ....' all the time.

[Ryan]

You were correct. I ran the 'unique' function that Stephen suggested and got [-1 0 1], so I just ran the im2bw command with a high threshold value and it solved the issue.

Answer 2

[Stephen]

I agree with Image Analyst. Is your image a jpg or has it been compressed? If so, there are little ghosts that lurk as low intensity values on certain pixels. bwlabel uses any value above 0 (or below?) so those sneaky pixels will get marked. Take a look at what a==0 looks like, or type unique(a) to see a list of the unique intensity values.