Kmeans vs imsegkmeans, what the difference?

Hi, Im currently working on image segmentation and was wondering what the difference bewtween the two functions is?

[Adam]

I imagine they use the same code under the hood, though each is only available in a specific toolbox and imsegkmeans looks like it is just a helpful wrapper function that does some of the work for you, mostly in terms of repeatability, judging from the top at the bottom. By default kmeans uses randomness in initialisation which means it will give different results each time, whereas imsegkmeans apparently does not do this and returns repeatable results each time. You can parameterise kmeans to do this also, it just isn't the default behaviour.

Answer 1

[Ganesh Regoti]

As per my understanding, there is not much difference in the underlying functionality of both kmeans and imsegkmeans as both does clustering of data.

imsegkmeans clusters the image based on the color of each pixel of that image.

kmeans clusters the data into given number of clusters. You can replicate the functionality of imsegkmeans with kmeans with some pre-processing.

You can refer following links to understand more about them

https://www.mathworks.com/help/stats/kmeans.html

https://www.mathworks.com/help/images/ref/imsegkmeans.html

https://www.mathworks.com/help/images/color-based-segmentation-using-k-means-clustering.html

[Tao Hu]

You said "You can replicate the functionality of imsegkmeans with kmeans with some pre-processing." Can you please specify what "pre-processing" should be done in order to make kmeans works the same as imsegkmeans? I do find that the result of kmeans contains a lot of noise unlike that of imsegkmeans. I really want to implement image segmentation using kmeans instead of imsegkeans because kmeans gives me more flexibility, fore example, I can segment a subregion (of arbitrary shape) of an image using kmeans, but not with imgsegkmeans, which requires the input must be an image (an N by M matrix).