Solving K-means issue

Hi,

I'm trying to use K-means to segment and image and then working on the resulting mask to clean it up, but K-means doesn't seem to give a fixed output everytime the script is executed so I'm not able to work on the resulting images correctly. How do I ensure that the output remains the same everytime ?

Answer 1

[Image Analyst]

It should, unless you've varied something, like the image you're processing, or the initial set of means. Does your script ask the user for any input (like filename, etc.), or use any random numbers?

[Lab Rat]

No, I'm not changing anything. Not even the input image. It just keeps changing when I run the script. . The final mask keeps interchanging between black and white. Sometimes the foreground is white while at other times it is black with the same thing with the background.

[Image Analyst]

Are you using some code or function from the Statistics Toolbox? I don't have that but if you edit your question to include it then maybe someone can reproduce it. Use a standard demo image that everybody has (because it ships with MATLAB) if you can.

Answer 2

[Walter Roberson]

The k-means algorithm involves random cluster initialization unless you provide initial cluster centers.