How to cluster the intensities in a gray scale image?

Dear all,

I have the following 8 bit gray image that contains 5 intensity levels .

Level-1 (black) = 159567 pixels

Level-2 = 68245 pixels

Level-3 = 33807 pixels

Level-4 = 10736 pixels

Level-5 (white) = 5736 pixels

However, the image I get is not always perfect and it looks something like what I show below.

Now, I want to know the number of pixels in each gray level in the noisy image (5 levels). Any idea how to find that?

I think I need a method that at first cluster the noisy image into 5 gray levels, and then find the number of pixels in each level (cluster).

K-means could be a good way to do that, but it always gives different result, so is there a better method than k-means? or is there a way to make k-means provide a correct and fixed answer?

Any idea will be appreciated.

Meshoo

Answer 1

[Sukuchha]

you can use, median filtering first and provide the resultant image to K-means. You would have use k-means X-times for reducing the effect of random generation of cluster centres.