Color Stealing Save
Adding color to binary fractals by stealing it
Color Stealing
Python implementation of the Color Stealing algorithm. In short, this technique constructs aesthetically pleasing fractals by copying color patterns from a given auxiliary image to a binary fractal.
Table of contents
Summary
Perhaps the most simple method to construct a fractal image is the Chaos Game. A fractal can be defined by its iterated function system (IFS), which is a set of n functions. For images, each function is with a corresponding probability
. The Chaos Game begins by sampling a random 2d coordinate
. At the j-th iteration, an integer
is sampled based on the distribution p. The next coordinate is computed as
. This coordinate is then plotted on a 2d grid, which was initially empty. After a number of iterations, the result is a binary image. Examples of such images can be found in the second column of the generated samples as well as at
templates/images.
The Color Stealing algorithm extends the Chaos Game by adding color to the constructed fractal. This requires an auxiliary image as input. Specifically, two IFSs are kept: one with functions and probabilities
for the fractal image and one with functions
and no probabilities for the auxiliary image. Similarly, two coordinates are initially sampled:
. At the j-th iteration, an integer
is likewise sampled from the distribution p, which is followed by calculations
and
. Then the fractal is extended:
. This process eventually results in images such as the third column of generated samples.
Generated samples
Below is a list of examples constructed by the provided scripts. Each sample was generated using the command python run.py -s -i 100000 -nt -l <ifs_path> -c <img_path>, where <ifs_path> was chosen from templates/ifs and <img_path> from assets/auxiliary. Running this commands produces a different result each time, due to rng. Obviously, all the examples are cherry-picked.
| Auxiliary Image | Binary Fractal | Color Stealing |
|---|---|---|
 |
 |
 |
|
 |
 |
|
 |
 |
 |
 |
 |
|
 |
 |
 |
 |
 |
|
|
 |
 |
 |
 |
 |
 |
 |
 |
|
 |
Installation
git clone https://github.com/davidsvy/color-stealing
cd color-stealing
pip install -r requirements.txt
Usage
python run.py OPTIONS
Saving and rendering:
-s, --save [FILE_NAME] If provided, saves the image and the IFS.
If FILE_NAME is not given, the image will
be saved at DIR/images/TIMESTEP.png and
the IFS at DIR/ifs/TIMESTEP.csv. DIR is
the next argument and TIMESTEP is the
timestep at the time of saving the files. If
FILE_NAME is given, the image will be saved at
DIR/images/FILE_NAME.png an the IFS at
DIR/ifs/FILE_NAME.csv. For the script to run,
either -p or -s must be provided.
-d, --dir DIR Directory where the image & the IFS file will
be saved.
-p, --plot If given, the constructed fractal will be
rendered using plt.imshow(). For the script
to run, either -p or -s must be provided.
-cm, --cmap CMAP cmap for for plt.imshow(). Accepted values are
'i', 'p', 'm', 'c', 'v'. Default value is 'i'.
CMAP is ignored if either -f or -c is provided.
Fractal parameters:
-r, --resolution RES Resolution of the square fractal image. Default
value is 1024.
-b, --batch_size BS Number of points calculated at each iteration.
If -f is given, setting BS <= 4 might result in
better visuals. Default value is 128.
-i, --n_iter N_ITER Number of iterations for the chaos game algorithm.
Default value is 50000.
-ni, --n_ignore N_IGNORE Number of initial iterations that will not be
rendered. Default value is 200.
-nt, --no_tqdm If given, the tqdm progress bar will not be used.
Loading saved IFS:
-l, --load PATH If not provided, the scripts will sample a random IFS.
If provided, the fractal stored inside PATH will be
generated. PATH must be a path to a valid IFS csv
file.
-m, --mutate PATH Same as the previous argument, with the difference
that the loaded IFS will be slightly mutated in a
random fashion.
Color Stealing:
-c, --color_steal PATH If not provided (and -f not provided), a binary
fractal will be generated. If provided, the fractal
will be constructed by copying color patterns from
the image stored at PATH using the Color Stealing
algorithm.
Fractal Flame:
-f, --flame If not provided (and -c not provided), a binary
fractal will be generated. If provided, the image
will be generated using the Fractal Flame
algorithm. Unfortunately, this mechanic is not
working well yet. Setting BS <= 4 might result in
better results.
-g, --gamma GAMMA Gamma parameter for the fractal frame algorithm.
Default value is 20.
-su, --sup SUP Supersampling ratio for the fractal flame algorithm.
Default value is 3.
Contact
The author of this repo can be mailed at [email protected].
Citations
@inproceedings{Barnsley2003ERGODICT,
title={ERGODIC THEORY , FRACTAL TOPS AND COLOUR},
author={Michael F. Barnsley},
year={2003}
}
Open Source Agenda Badge
