K-means-Image-Compression

I have implemented an unsupervised image compression program using K-means clustering. The program is to be given an array of clusters, depending on the values of the clusters respective K-means algorithm is ran and k centroids are calculated on which colors are mapped.
Reducing the number of different colors reduce the size of the image a lot as the pixels can be mapped as a vector corresponding to a single RGB value rather than all pixels having unique values.
I have used skimage for image input output & sklearn.cluster for kmeans algorithm.

To run the program following dependencies have to be installed;

pip install sklearn
pip install skimage

Make sure the correct path is added where you are placing your image files.

I used cluster values 2,3,5,10,15 and 20 and got the following results.

for k = 2;

for k = 3;

for k = 5;

for k = 10;

for k = 15;

for k = 20;

Original image;

Observation;

What we can observe is that with k = 2 we have very little information of features; we can only observe eyes and the shape of the face as we move to k= 3; we observe all the features and can predict the gender of the person accurately; then on k=5 we get a cartoony image of the person which is a very popular filter nowadays. At k =10 we get the image which will get as close to original with the least size with all features and smooth edges which gets better at k = 15 where we can observe easily the wave of different centroid on the skin of the person. At k=20 we have a compressed image close to the original with a fraction of the size.