can some help me please with Otsu thresholding matlab code in which the dimensions, width, height, of the output image is equal to the orignal image (input image). Thanks in advance.

[Walter Roberson]

please post your code

[Arshad Ali]

I=imread('1 (4).png');
figure(1),imshow(I);
figure(2),imhist(I);
n=imhist(I);
N=sum(n);
max=0;
for i=1:256
   P(i)=n(i)/N;
end
  for T=2:255 
     w0=sum(P(1:T));
    w1=sum(P(T+1:256));
     u0=dot([0:T-1],P(1:T))/w0;
      u1=dot([T:255],P(T+1:256))/w1;
      sigma=w0*w1*((u1-u0)^2);
      if sigma>max
          max=sigma;
          threshold=T-1;
      end
  end
     bw=im2bw(I,threshold/255);
     figure(3),imshow(bw);

[Arshad Ali]

i applied this code it gives the accurate segmentation but the problem is it can change the width, height and dimension of the output image. i want Otsu threshold code in which the width, height and dimension remain same with the input image. Thanks in advance.

Hope you will help me with that sir.

[Walter Roberson]

No, in that code, bw is certain to be exactly the same height and width as the original image, I. However, bw will be black and white (2D) instead of color (3D). If you need the exact same output dimensions then

I_out = repmat( im2uint8(bw), [1 1 3]);

[Arshad Ali]

Hello Sir. whether i can use this command i try it but did not work same problem.

[Arshad Ali]

This is the output image which i get but if you check the both image properties its dimension, width and height is different i want it same for furthur investigation.

[Image Analyst]

The deprecated im2bw() does not change the number of rows and columns in the image. What is the size you say it's giving and what size do you think it should be?

[Arshad Ali]

but i am working on otsu algorithm only

[Walter Roberson]

it looks to me as if you are using saveas or print or getframe on the displayed image and then complaining that it is not the same size as the original .

[Image Analyst]

Good guess Walter - that's probably it. When I use "whos" to look at the I and bw variable:

  Name        Size             Bytes  Class      Attributes
  bw        256x320            81920  logical 
  Name        Size             Bytes  Class    Attributes
  I         256x320x3          245760 uint8       

You can see that they're both 256 by 320 - the same lateral size (rows and columns), NOT different.

Arshad, WHY do you say they're of a different size? How did you check that? Please supply proof.

[Arshad Ali]

its look perfect in matlab but when i save the output image from matlab and then check both images properties its different even i am not using save as.

[Arshad Ali]

[Arshad Ali]

when i save the output image from matlab. There is a white margin around the image.

[Walter Roberson]

how are you saving the output image ?

[Arshad Ali]

i found my mistake instead of using imwrite() command i save images directly . when i used imwrite() command it save image with original dimension .. Thank you Walter Roberson. Thank you so much..

Answer 1

[KALYAN ACHARJYA]

Hello Arshad, otsu thresolding techniqie is very popular thresholding approach in image segmentation domain. I have posted the following code from Gonzalez book (Digital Image Processing using Matlab), requested you to refer the book for detail explanation.

function [T,SM]=otsuthresh(h);
h=h/sum(h);
h=h(:);
i=(1:numel(h))';
P1=cumsum(h);
m=cumsum(i.*h);
mG=m(end);
sigSquared=((mG*P1-m).^2)./(P1.*(1-P1)+eps);
maxSigsq=max(sigSquared);
T=mean(find(sigSquared==maxSigsq));
T=(T-1)/(numel(h)-1);
SM=maxSigsq/(sum(((i-mG).^2).*h)+eps);
end