Appendices

Colour classification

Cluster analysis can also be performed on the RGB values of the image. In doing so, we can essentially group pixels into similar categories and then re-visualise the image using the average value of these groups. In doing this, we can simulate the image over different levels of ‘colour resolution’.

The series of images below show from top left to bottom right:

1. The original image
   
2. 2 colour clustering
   
3. 3 colour clustering
   
4. 4 colour clustering
   
5. 5 colour clustering
   
6. 10 colour clustering
   
7. 15 colour clustering
   
8. 20 colour clustering
   
9. 25 colour clustering

What I find interesting about this clustering of pixel colours is that you can segment an image to any number of distinct colours, gradually reassembling it as you increase the number of clusters.

Colour variation

I can also look at groups of similar images and compare their two dimensional colour space plots:

As expected, these four images of volleyballers serving, returned very similar colour spaces, made up mostly of skin tones and blues from the sky.

This second example of beaches (all taken on the same day) show more variation in colour on the principal component plots, the dominant colours of ocean and sky are evident throughout all three.