deletions | additions       diff --git a/We_segmented_the_images_using__.tex b/We_segmented_the_images_using__.tex  index 8cfcaac..f058a83 100644  --- a/We_segmented_the_images_using__.tex  +++ b/We_segmented_the_images_using__.tex 
...
We segmented the images using the K-Means algorithm with different values of $k$. Ideally, we want a $k$ that would cluster the images such that the clusters contain either infected pixels only or healthy pixels only. For this experiment, we tried setting $k$ to 2, 3, and 4. Figure \ref{fig:kImageSegment} shows the results of the segmentation. When $k$ is set to 2, it is expected that one cluster will contain the infected pixels and the other will contain  the image healthy pixels. But in our data set, most of the images cannot be separated well using just two clusters. Figure (\ref{fig:kImageSegment}b)  is an example of this. Its healthy pixels had a yellow green color while its infected pixels had parts which are dark brown and parts which are light brown, causing  the image one cluster to have mix healthy and infected pixels. Setting $k$ to 3 significantly improves the results. Increasing $k$ to 4 that all the clusters does not have ... Based on the criteria, a bigger $k$ would give better results since the variance within clusters decreases. However, more clusters would mean more computational complexity and longer processing times.