can i use k-means algorithm for segmenting the cell nucleus and cytoplasm?

i have used k-means clustering algorithm for segmenting cells but it doesn't. i didn't know about k-means algorithm in detail. i don't know whether this algorithm suits or not. suggest me a way. i have attached my image that i have been working. the inner dark region is nucleus and outer region is cytoplasm.

<<

>>

Answer 1

[Image Analyst]

This will walk you through it: http://www.mathworks.com/products/demos/image/color_seg_k/ipexhistology.html

[sou]

i have tried this sir.. but did doesnt works out.. suggest me a way...

[Image Analyst]

You can use thresholding in HSV color space, plus a few morhpological functions to clean up small noise regions. See my File Exchange for a demo.

[sou]

from were should i refer your file exchange.. can you please send me the link sir

[Image Analyst]

http://www.mathworks.com/matlabcentral/fileexchange/?term=authorid%3A31862

[sou]

it doesnt seperate nucleus and cytoplasm sir

[Braiki Marwa]

Svp j'ai utilisé cet algorithme mais il me renvoie toujours erreur!!!!!!!!!!!!!!!!!!

[Image Analyst]

sou, then you need to improve the color segmentation algorithm.

Braiki, I'm not sure what you said. Something about errors, but you didn't provide the errors or the code.

Answer 2

[Alex Taylor]

In this case, I found that I was able to get a reasonably good segmentation of the nucleus by working directly in the RGB colorspace instead of LAB as is done in the example:

%%Read and display input image

A = imread('http://www.mathworks.com/matlabcentral/answers/uploaded_files/26706/inter4.JPG');
A = im2double(A);
imshow(A)

numRows = size(A,1);
numCols = size(A,2);

numPoints = numRows*numCols;
X = reshape(A,numRows*numCols,[]);

Normalize features to be zero mean, unit variance

X = bsxfun(@minus, X, mean(X));
X = bsxfun(@rdivide,X,std(X));

%%Classify color features using kmeans
% Repeat k-means clustering five times to avoid local minima when searching
% for means that minimize objective function. The only prior information
% assumed in this example is how many distinct regions of texture are
% present in the image being segmented. There are two distinct regions in
% this case.
L = kmeans(X,3,'Replicates',5);

%%Visualize segmentation using |label2rgb|
L = reshape(L,[numRows numCols]);
figure
imshow(label2rgb(L))

%%Visualize segmented image using |imshowpair|
% Use imshowpair to examine the foreground and background images that
% result from the mask BW that is associated with the label matrix L.
Aseg1 = zeros(size(A),'like',A);
Aseg2 = zeros(size(A),'like',A);
BW = L == 2;
BW = repmat(BW,[1 1 3]);
Aseg1(BW) = A(BW);
Aseg2(~BW) = A(~BW);
figure
imshowpair(Aseg1,Aseg2,'montage');