Labelling image

Hello.

I`ve been trying to recognise simple shapes on photos and I`m not allowed to use bwlabel function, so I had to write simple labelling algorithm. It`s based on the one in: Tadeusiewicz R., Korohoda P. "Computer Analysis and Image Processing".

I don`t know the proper name for it in english, but I think it`s called: Connected-component labeling.

It works for basic example (pic1) but when there`s a lot of lines like this -> /, sudennly algorithm starts to assign few colors to one object (pic2 and pic3). bwlabel functions works properly of course.

close all;
clear all;

%%OBRAZKI

%%O = imread('rys1.png'); %%correct
%%O = imread('rys2.png'); %%bad
O = imread('rys3.png'); %%bad

%%binarisation
k = 80;
O = im2bw(O, k/255);
O=~O;
figure(1);
imshow(O);

%%algo
[X Y] = size(O);
OW = zeros(X, Y);
L = 1;
tab = zeros(255);

for i=2:X-1
    for j=2:Y-1
        if O(i, j) == 1
            s = [OW(i-1, j-1), OW(i-1, j), OW(i-1, j+1), OW(i, j-1)];
            if max(s) == 0
                OW(i, j) = L;
                tab(L) = L;
                L = L + 1;
            elseif min(nonzeros(s)) == max(nonzeros(s))
                OW(i, j) = min(nonzeros(s));
            else
                OW(i, j) = min(nonzeros(s));
                tab(max(nonzeros(s))) = min(nonzeros(s));
            end
        end
    end
end

for i=2:X-1
    for j=2:Y-1
       if OW(i,j) ~= 0
           OW(i,j) = tab(OW(i,j));
       end
    end
end

figure(2);
subplot(1,3,1);
imshow(OW, []);
title('Reczna');

OW = bwlabel(O, 4);
subplot(1,3,2);
imshow(OW, []);
title('BWLABEL 4');

OW = bwlabel(O, 8);
subplot(1,3,3);
imshow(OW, []);
title('BWLABEL 8');

Pictures are here:

http://i.imgur.com/qBnoY.png pic1

http://i.imgur.com/a9N4Q.png pic2

http://i.imgur.com/uXhiA.png pic3

Forgive me that you have to change the names yourself.

I would appreciate any suggestions or remarks.

Sincerely, Muss

PS I know my english is terrible, I`m so sorry.

Answer 1

[Image Analyst]

It looks like, from this line:

s = [OW(i-1, j-1), OW(i-1, j), OW(i-1, j+1), OW(i, j-1)];

that you have some kind of weird pattern for doing connected components labeling. Instead of a cross for 4-connected

0 1 0
1 1 1
0 1 0

it's using an "L" shape:

1 1 0
1 0 0
1 0 0

Instead of just looking at those 4 neighbors, look at all 8, so s will have additional terms to capture the corners:

s = [OW(i-1, j-1), OW(i-1, j), OW(i-1, j+1),... OW(i, j-1), OW(i, j), OW(i, j+1), ... OW(i+1, j-1), OW(i+1, j), OW(i+1, j+1)];

so you get an 8-connected pattern:

1 1 1
1 0 1
1 1 1

[Muss]

Well, algorithm goes from left to right every row, so why I should check pixels like this?

1 1 1

1 0 1

1 1 1

According to this - http://en.wikipedia.org/wiki/Connected-component_labeling, I should use this L shape (called 8-connectivity):

1 1 1

1 0

Answer 2

[Muss]

Fixed - a bit. There is still something wrong, but:

tab(max(nonzeros(s))) = min(nonzeros(s));

change it to:

tab(max(nonzeros(s))) = tab(min(nonzeros(s)));

It will make the algorithm little bit more resistant to those little stairs where the color used to change.