Rearrangements

Making One Image From the Pixels of Another

Introduction

My idea was to take the pixels of one famous painting and rearrange them to try to form the image of another famous painting. The pixels cannot be altered in any way, and each pixel has to be used exactly once. The concept notwithstanding, the results were not as interesting as I'd hoped. They mostly look like strange colorizations of the target image.

Examples

Source Pixels	Target Image	Resulting Rearrangment
Van Gogh's Starry Night	Leonardo da Vinci's Mona Lisa
Edvard Munch's The Scream	Leonardo da Vinci's Mona Lisa
Picasso's Three Musicians	Monet's Woman, Umbrella Turned to the Left
Picasso's Three Musicians	Tommie Smith and John Carlos at the 1968 Summer Olympics
Richard Diebenkorn's Cityscape I	Gustave Caillebotte's Paris: A Rainy Day

How To Make Them

The process of making these is pretty much the same as making a photomosaic except that the source images are all 1x1 pixel. At first I thought I'd need to split each soure image into n 1x1 images and then use a photomosaic tool, but as it turns out metapixel already supports this:

metapixel --metapixel target_image.jpg output_image.png -x source_image.jpg -w 1 -h 1 --search global

-x causes the given image to be used as the source for the tiles.
-w and -h set the size of the tiles to 1x1.
--search global causes no pixel to be used more than once.

Of course, the source and target images must have the exact same number of pixels for this to work. That was accomplished easily enough with netpbm:

djpeg full_size_image.jpg | pamscale -xsize 63 -ysize 84 | cjpeg > small_image.jpg

The hard part was finding a set of dimensions for the source and target that gave the same number of pixels. Most of the source images need some distortion to accomplish this.

The problem with metapixel is that it uses a lot of memory when called with --search global -w 1 -h 1. The tiny examples above consumed 500MB of memory on each run!

Chez Zeus:Rearrangements