Hi,
For data in matrix form, the kmeans algorithm assumes that individual rows are the data instances. So when we apply kmeans on an N*N matrix, it returns a N*1 index array instead of an N^2*1 index array.
As a workaround, we can consider
- Converting the Ks matrix into a column vector Ks_vec with the same number of elements as Ks.
- Applying kmeans on Ks_vec to obtain the labels for each element,
- Reshape it back to the shape of Ks.
grayImage= imread('CYST RENAL -87.jpg');
g = rgb2gray(grayImage);
g = double(g);
sigma = 0.4;
[n,d] = size(g);
nms = sum(g'.^2);
Ks = exp(-(nms'*ones(1,n) -ones(n,1)*nms + 2*g*g')/(2*sigma^2));
Ks_vec = reshape(Ks,[size(Ks,1)*size(Ks,2),1]);
[m n]=kmeans(Ks_vec,3);
m=reshape(m,size(Ks,1),size(Ks,2));
B=labeloverlay(Ks,m);
figure;
imshow(B);
Hope this helps!
