K-means Clustering for Color Grouping

I need to cluster different colors using k-means clustering algorithm.

when i use function like this with k=4 for 5 colors image.

[cluster_idx cluster_centre]=kmeans(inputImg,k,'distance','sqEuclidean','Replicates',3);

It will end up with empty cluster at iteration 1 and no partition. If i run multiple times,then at one time it will partition the colors .

then if i change above function like this [cluster_idx cluster_centre]=kmeans(inputImg,k); and then also end up with same result.

Why it showing like this even though image is having 5 colors..? How to debug this problem..? Does any one know please help me out.

Thank You.

Answer 1

[Image Analyst]

Have you seen the Mathworks demo: http://www.mathworks.com/products/demos/image/color_seg_k/ipexhistology.html. I don't have the stats toolbox so I can't help you debug it anymore.

[Chandra Shekhar]

I used same demo code, but still i am getting error.

In my program, i used color reduction image as a input to the 'kmeans'function.In that the image is having 5 colors and i need to group those colors by choosing k=4 or 5;

In demo program the direct input image is used as input to the kmeans function.

Does any one know please help me..

Thank You.

[Image Analyst]

I can't help you because I don't have the stats toolbox. Perhaps you can call the Mathworks and ask why their demo returns 4 clusters (or 4 plus an empty 5th) when you're specifying 5 clusters.