Image + K means clustering + Convex Hull

ExperimentWithConvexHull.R

library(magick)
library(imager)
library(tidyverse)


create_pebble_art <- function(img_url,cluster_cnt=500,samp=10000){
  
  img <- image_read(img_url)
  
  ## convert image to df
  im.df.colour <- img %>%
    image_resize(400)%>%
    image_quantize() %>%
    magick2cimg() %>%
    as.data.frame(wide="c") %>% ## so that rgb value is in separate column.
    mutate(hex=rgb(c.1,c.2,c.3)) 
  
  ## original image is too big, so sample them...
  im.df.mini <- im.df.colour %>%
    sample_n(size=samp)
  
  ## run K-means so that group is formed on distance and colour
  k_res <-im.df.mini %>% select_if(is.numeric) %>% kmeans(centers=cluster_cnt)
  
  im.df.mini$grp = k_res$cluster
  
  
  im_simp <-im.df.mini %>% 
    select(x,y,grp) %>%
    group_by(grp) %>%
    nest() %>%
    mutate(hull=map(data,chull)) %>%
    mutate(out=map2(data,hull,~.x[.y,])) %>%
    unnest(out) %>%
    left_join(im.df.colour %>% select(x,y,hex))
  
  p <-im_simp %>% 
    add_count(grp) %>%
    filter(n>2) %>%
    ggplot() +
    theme_void() +
    scale_y_reverse() +
    geom_bspline_closed(aes(x=x,y=y,fill=hex,group=grp), alpha=0.8, size=0.1)+
    scale_fill_identity() +
    scale_color_identity() 
  
  print(p)
  
}


create_pebble_art(img_url="https://live.staticflickr.com/65535/49664808511_eec5b4e36b_z.jpg",100)
create_pebble_art(img_url="https://live.staticflickr.com/65535/49664808511_eec5b4e36b_z.jpg",200)