Utilizing K-means to extract colours from your favourite images

November 11, 2018
By

[This article was first published on R on Chi's Impe[r]fect Blog, and kindly contributed to R-bloggers]. (You can report issue about the content on this page here)
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.




I have been playing with package called imager, documentation on this package was extremely helpful! I have read through “getting started” as well as few other tutorial & examples.

I love colours… Tools like colourlovers, Adobe Color CC, Canva Color Palette Generators are great extracting colours from photo (or in general just getting inspirations for colour palette), but I wanted to figure out the way to extract colours from image using R.

Preps

Loading up packages & loading up image to R, and extracting RGB info out of image. First I need to load up image. I’m using below abstract image with bunch of different colours that I’ve created with Photoshop just for fun.

library(tidyverse) ##
library(imager) ## such a fun package.  I want to learn more.
library(treemap) ## 
library(ggvoronoi) ## because I'm currently addicted to voronoi diagram.

## Load up the image using load.image function! 
im < - load.image("https://farm4.staticflickr.com/3316/3333507738_9d36d39f6d_b.jpg") ## colourful abstract image. 
#im <- load.image("https://farm9.staticflickr.com/8125/8659010017_54a885f12a_z.jpg") ## mainly blue
#im <- load.image("https://farm2.staticflickr.com/1939/30915465767_2d9a733510_z.jpg") ## animoji! 

## View the image I just loaded. 
plot(im, main="Original Image I Want To Get Some Colours Out Of!")

df_size < -dim(im)[1]*dim(im)[2]  ## numbers of row I'd get if i use it as it is...
max_row_num <- 150000 ## number of maximum row I want (just to limit my size!)

## If df is too big, it's too slow to process on my computer, so shrink the image
shrink_ratio  <- if(df_size > max_row_num) {max_row_num / df_size } else {1}
im < - im %>% imresize(shrink_ratio)
# plot(im) if you want to check how the image has been resized.

## get RGB value at each pixel
im_rgb < - im %>% 
  as.data.frame(wide="c") %>%
  rename(red=c.1,green=c.2,blue=c.3) %>%
  mutate(hexvalue = rgb(red,green,blue)) ## you can create hexvalue using red, green blue value!

## turn image into Grayscale and get luminance "value" too. 
im_gray < - im %>%
  grayscale() %>%
  as.data.frame() 

## combine RGB info and Luminance Value Dataset together.
im_df < - im_rgb %>% 
  inner_join(im_gray) 

im_df %>% head()
##   x y         red     green      blue hexvalue     value
## 1 1 1 0.001528153 0.6012714 0.7346048  #0099BB 0.4360151
## 2 2 1 0.003336439 0.5943815 0.7305651  #0198BA 0.4320482
## 3 3 1 0.005112951 0.6265604 0.7657791  #01A0C3 0.4554402
## 4 4 1 0.005520480 0.6726386 0.8097380  #01ACCE 0.4875841
## 5 5 1 0.011926270 0.7021898 0.8367487  #03B3D5 0.5099122
## 6 6 1 0.011128983 0.6695389 0.8149510  #03ABD0 0.4880113

Applying K-Means Algorithm on RGB value from Image

Using kmeans function is pretty simple, I’m selecting 12 as k in below example, simply because I wanted to get 12 distinct colours from the picture.

## Pick k value to run kMean althorithm.
## But to extract colours, I'd pick k as number I want back! 
my_k < - 12

## Running kmeans algorithm on red, green and blue value to gather similar colour together
kmean_rgb <- kmeans(im_df %>% select(red,green,blue), centers=my_k)

## append cluster id to im_df datasets.
im_df$cluster_num < - kmean_rgb$cluster

## center values can be used as cluster colour! 
kmean_center <- kmean_rgb$centers %>% as.data.frame() %>% 
  mutate(group_hex = rgb(red,green,blue), cluster_num = row_number()) %>%
  inner_join(im_df %>% count(cluster_num))

## I can also save the colour palette for future use as well.
my_colour < - kmean_center$group_hex
my_colour
##  [1] "#12AAD8" "#F178C3" "#E9E0CE" "#DF3658" "#933DB6" "#659634" "#E62AB1"
##  [8] "#73D2B0" "#F5E86F" "#EAC547" "#4363C2" "#93D55F"

Viewing Colour Palettes

I wanted to view the colour palette more visually, instead of HexValues….

num_row = floor(my_k/4)
kmean_center %>% 
  ggplot(aes(x=(cluster_num-1)%%num_row,y=floor((cluster_num-1)/num_row))) + 
  geom_tile(aes(fill=group_hex)) +
  geom_label(aes(label=paste(cluster_num,":",group_hex,"\n",n,"distinct hexvalues")), 
             family="Roboto Condensed", lineheight=0.8) +
  scale_fill_manual(values=sort(kmean_center$group_hex), guide="none") +
  theme_void(base_family="Roboto Condensed") +
  labs(subtitle=paste0("k-Mean clustering center colour with ", my_k, " clusters" )) +
  scale_y_reverse()

Which colours get clustered together?

I wanted to see which colours were bundled together in same cluster. I took a sample because plotting more colours simply just took way too long on my machine…

im_df %>% 
  sample_n(size=10000) %>% ## I'm just going to take sample to make the drawing bit faster...
  count(cluster_num,hexvalue) %>%
  treemap(index=c("cluster_num","hexvalue"),
          type="color",
          vSize="n", 
          vColor="hexvalue",
          algorithm = "squarified",
          fontfamily.labels="Roboto Condensed", 
          fontfamily.title="Roboto Condensed",
          border.col=c("#ffffff","#ffffff50"),
          fontsize.labels=c(24,0),
          aspRatio=16/9,
          title="Clustering with RGB")

Bonus: Voronoi Abstract Art

Below is more fun to do on other images, but I wanted to just use ggvoronoi package because I’m currently in love

Looks like stained glass! 🙂

## Create Mini sets, choose to select grayscale value that are bigger more often.
im_df_mini < -im_df %>%
  sample_n(size=1000, weight=(1-value)) 

im_df_mini %>%
  ggplot(aes(x=x,y=y)) +
  geom_voronoi(aes(fill=hexvalue), color="#000000", size=0.1) +
  scale_fill_manual(values=sort(unique(im_df_mini$hexvalue)), guide="none") +
  theme_void() +
  scale_y_reverse() +
  coord_fixed()

Using Plotly package

I’ve always been curious to use plotly, but I haven’t had chance to yet. Since RGB values are 3 different values, I wanted to use xyz axis to plot (and I didn’t know how to do that in ggplot2), so here’s yet another way to view how the colour got clustered more visually.

library(plotly)
im_df %>% sample_n(size=2000) %>% 
  plot_ly(x = ~red, y=~blue, z=~green, color=~cluster_num, colors=my_colour, size=~value) %>%
  add_markers()   

To leave a comment for the author, please follow the link and comment on their blog: R on Chi's Impe[r]fect Blog.

R-bloggers.com offers daily e-mail updates about R news and tutorials about learning R and many other topics. Click here if you're looking to post or find an R/data-science job.
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.



If you got this far, why not subscribe for updates from the site? Choose your flavor: e-mail, twitter, RSS, or facebook...

Comments are closed.

Search R-bloggers

Sponsors

Never miss an update!
Subscribe to R-bloggers to receive
e-mails with the latest R posts.
(You will not see this message again.)

Click here to close (This popup will not appear again)