CODEWITHSUNDEEP
rimageneural-network
Rodrigo Stv
Rodrigo Stv

Reputation: 425

Generate pictures in R for use in training algorithms

I want to train a neural network to count circles on an image by generating thousands of examples. So I need a function to generate a picture with a random number of circles in it.

I have been playing with plots but I guess there must be a better way directly manipulating pixel matrices. Any ideas?

Upvotes: 1

Views: 96

Answers (2)

aoles
aoles

Reputation: 1545

You can use the Bioconductor package EBImage to draw circles on pixel matrices and save them as PNG files, as demonstrated below.

library("EBImage")

set.seed(1)

images = 10   # number of images to generate
size = 500    # X-Y dimensions of images
circles = 50  # max number of circles
rmin = 5
rmax = 25     # max circle radius

xy = size - 2*rmax
rr = rmax - rmin
n = sample.int(100, images)

for (i in 1:length(n)) {
  img = Image(1L, c(size, size))
  x = sample.int(xy, n[i]) + rmax
  y = sample.int(xy, n[i]) + rmax
  r = sample.int(rr, n[i], replace = TRUE)
  for (k in 1:n[i]) {
    img = drawCircle(img, x[k], y[k], r[k], col=0)
  }
  writeImage(img, sprintf("img%.5d.png", i))
}

img00001.png

In case you need filled circles, you can provide the argument fill = TRUE to the function drawCircle. You can also easily change the image background in the call to Image, or vary the circle color with the argument col to drawCircle. For example, use 

...
img = Image(0L, c(size, size))
...
img = drawCircle(img, x[k], y[k], r[k], fill = TRUE, col = runif(1, .1))
...

for grey circles on black background.

enter image description here

Upvotes: 1

IRTFM
IRTFM

Reputation: 263382

Might as well throw it out as an answer:

set.seed(123)
n= sample(500, 1)  # picks 1 number in range of 1 to 500
plot( runif(n), runif(n), cex=runif(n,1,6),  # Varies size and location
                         main=bquote( .(n)~ Random~Circles) )

enter image description here

Upvotes: 1

Related Questions