problem in seeded region growing algorithm

I am trying to implement seed region growing algorithm but its not giving proper output.Please help me solve the error..

here is my code..

clc;clear all;close all;

r=imread('e:\IMAGES\pears.jpg'); % read the image

[rows columns numberOfColorChannels] = size(r);

if numberOfColorChannels > 1
  r = rgb2gray(r); % converting image to grayscale
else
  r = r; % It's already gray.

end
[row col]=size(r)
g=zeros(row,col);
I=zeros(size(r));

I=r;

S=zeros(size(r));

S(80:82,172:174)=ones(3,3); %initial seed image

figure,imshow(S),title('Initial Seed Image')

T=50; %threshold value

r=double(r);
w = [1 1 1;1 1 1;1 1 1 ];  
SI=size(r)
p=size(r);
for x=2:1:p(1)-1
  for y=2:1:p(2)-1  
      a1=[w(1)*S(x-1,y-1) w(2)*S(x-1,y) w(3)*S(x-1,y+1) w(4)*S(x,y-1) w(5)*S(x,y) w(6)*S(x,y+1) w(7)*S(x+1,y-1) w(8)*S(x+1,y) w(9)*S(x+1,y+1)];
      AI(x,y)=min(a1);
  end
end
SI= padarray(AI,[1 1],'symmetric','post')

SI=uint8(SI)

J=find(SI); 

S1=r(J);  % seed value obtained from SI

seedvalue=S1;

S=abs(r- seedvalue)<= T; % thresholding the absolute difference between original image & seed value

figure,imshow(S),title('thresholding output')

      for p=1:1:row
      for q=1:1:col
          if(SI(p,q)==1)
              for n=q:1:col
              if S(p,q)==S(p,n)
                  g(p,n)=1;
              end
              end
              for n=q:(-1):1
              if S(p,q)==S(p,n)
                  g(p,n)=1;
              end
              end
              for n=p:-1:1
              if S(p,q)==S(n,q)
                  g(n,q)=1;
              end
              end
              for n=p:1:row
              if S(p,q)==S(n,q)
                  g(n,q)=1;
              end
              end
             for n1=p:1:row
               for  n2=q:1:col
              if S(p,q)==S(n1,n2)
                  g(n1,n2)=1;
              end
              end
              end
              for n1=p:1:row
               for   n2=q:(-1):1
              if S(p,q)==S(n1,n2)
                  g(n1,n2)=1;
              end
              end
              end
              for n1=p:-1:1
               for   n2=q:(-1):1
              if S(p,q)==S(n1,n2)
                  g(n1,n2)=1;
              end
              end
              end
                for n1=p:-1:1
                for  n2=q:1:col
              if S(p,q)==S(n1,n2)
                  g(n1,n2)=1;
                  end
              end
              end

              end
              end
          end

g=uint8(g)

SE = ones(1,1);
g =bwlabel(imdilate(g,SE),8);%reconstructing the image

%g=logical(g)

%I(g)=0

%figure,imshow(I),title('Final Output')
Lrgb = label2rgb(g);
figure, imshow(Lrgb)
title('output of region growing process')

figure, imshow(I), hold on
himage = imshow(Lrgb);
set(himage, 'AlphaData', 0.3);
title('final output superimposed on original image')

[Walter Roberson]

How does the output you are seeing differ from the output you expect? Is there an error message?

[Image Analyst]

I ran it and it looks like it ran fine to me. All the pears in figure2 that were touching the seed point and touching each other got selected as the same object in subsequent images, which is what it should do. So, what's the problem? You'd get the same thing just calling bwlabel() or bwconncomp() on your binary image.

[aarti sawant]

but the output must show only one pear of which seed point is selected.I dont want other pears to be highlighted.

[Image Analyst]

Well then your binary image is wrong, and that is calculated before you even start the region growing. All those other pears are connected to the seed pear in your binary image.

Anyway, if you have a seed point, you don't need to do all that stuff you did since there is a function in the Image Processing Toolbox that does all that for you in one line. It's called imreconstruct (). But of course, it also requires that you have the pears separated before calling it just like your algorithm does.

[aarti sawant]

i know imreconstruct command gives the expected output but is there any other way of getting correct output without using direct command

[Image Analyst]

You can write your own region growing, or labeling, code as you did. Are you not seeing how, in your binary image, that the vast majority of the pears in your image area connected by white pathways to the pear where you placed the seed? Please let me know if you realize that. And if you do, why you don't think your growing code will reach out to all of those?

[aarti sawant]

I am not getting what you are trying to say

[Image Analyst]

I'm not sure how else to say it. What do you think region growing does? It captures (labels) all pixels that are connected to the seed pixel according to some criteria, like they have the same or similar gray level. You have just a binary image. You have about 3 or 4 white binary blobs, not 2 dozen pears. Look at your binary image - see? Just 3 blobs. That one gigantic blob that takes up almost your whole image? It's right over your seed point so that whole gigantic blob is going to be selected. Why should it get just the pear over the seed? According to your binary image that you are using, that pear is physically connected to the next one, and the next one, and the next one after that. Virtually all of them are connected. I suggest you look at how you came up with that binary image. And I suggest you look at this example of marker controlled watershed segmentation, which luckily for you is done with the pears demo image. http://matlab.wikia.com/wiki/FAQ#After_installation.2C_MATLAB_crashes_or_gives_an_error_message_when_I_try_to_run_MATLAB.

[aarti sawant]

thank you Image Analyst

[Hanz Ero]

aarti sawant, can you help me, please give your code to me for the sake of college.

Answer 1

[renugadevi ramadoss]

i have a doubt in setting initial seed points using seed determination for segmentation.

[Walter Roberson]

Please explain your doubt further.