Problems with kmeans clustering

OI have used the following code to segment the attached image. But each iteration on the same image shows different result. How can i solve this by using rng('default'); ?

[Adam]

You should just need to explicitly set the seed (either to 'default' I guess or to any seed of your choice) before each call to kmeans if you want the same result each time.

[sam CP]

 %k-means clustering algorithm
imData = reshape(Y,[],1);
imData = double(imData);
[IDX nn] = kmeans(imData,'default');
imIDX = reshape(IDX,size(Y));
figure, imshow(imIDX,[]),title('Image after applying k-means Clustering Algorithm');

Where can i apply the rng('default'); ?

Answer 1

[the cyclist]

Looking at your code, you should be able to put the line

rng('default')

literally anywhere before the call to kmeans, because you do not call any other random number functions. But the safest bet might be to call it in the line just before the call to kmeans, in case you do something differently later.

But, also, I don't think you put 'default' in the actual kmeans call. So it should be like this ...

%k-means clustering algorithm
imData = reshape(Y,[],1);
imData = double(imData);
rng('default')
[IDX nn] = kmeans(imData);
imIDX = reshape(IDX,size(Y));
figure, imshow(imIDX,[]),title('Image after applying k-means Clustering Algorithm');

[sam CP]

Error using kmeans (line 155) At least two input arguments required.

Error in Untitled2 (line 49) [IDX nn] = kmeans(imData);

Is there again need to specify the cluster number after applying rng('default') this line into my code.

[Adam]

Your original (from your comment above)

[IDX nn] = kmeans(imData,'default');

is not valid syntax for calling kmeans as far as I can see. There is no function signature where 'default' makes sense as a 2nd argument.

The code you attached calls it totally differently so just use that exactly as you did before.

[sam CP]

Error using kmeans (line 265) K must be a positive integer value.

Error in Untitled2 (line 49) [IDX nn] = kmeans(imData,'default');

[Adam]

I;m really not understanding how your code relates in any way to the code you are saying creates an error. At a glance your code here looks fine, just add

rng('default')

on the line above

[cluster_idx, cluster_center] = kmeans(ab,nColors,'distance','sqEuclidean', ...
                                      'Replicates',2);

I don't know where:

[IDX nn] = kmeans(imData, 'default');

is coming from at all.

[sam CP]

The result will not be changed in each iteration on the same image. Thank you. The code works.

[sam CP]

What changes i have to make in the code when i want to make 4 clusters?

[Adam]

Change nColours to 4.

[sam CP]

it works

[Image Analyst]

Yeah, but let's put "works" in quotation marks because kmeans() is not a good method for finding brain tumors. Imagine what your algorithm would find for class 4 if there were no tumor present, or a very small one. Yeah, see what I mean?

[sam CP]

I have already found that the kmeans clustering method can't be detect the tumor when it is very small.