#quantization #palette #spatial #dither #halftone

bin+lib rscolorq

Spatial color quantization, a Rust port of scolorq

2 releases

0.1.1 Nov 16, 2020
0.1.0 Oct 23, 2020

#60 in Images

MIT/Apache

89KB
2K SLoC

rscolorq

Build Status Crates.io Docs.rs

dithered mountains

Rust port of Derrick Coetzee's scolorq, based on the 1998 paper "On spatial quantization of color images" by Jan Puzicha, Markus Held, Jens Ketterer, Joachim M. Buhmann, & Dieter Fellner. Spatial quantization is defined as simultaneously performing halftoning (dithering) and color quantization (limiting the colors in an image). For more information, visit the original implementation's website.

The algorithm is excellent for retaining image detail and minimizing visual distortions for color palettes in the neighborhood of 4, 8, or 16 colors, especially as the image size is reduced. It combines limiting the color palette and dithering the image into a simultaneous process as opposed to sequentially limiting the colors then dithering. Colors are chosen based on their context in the image, hence the "spatial" aspect of spatial color quantization. As in Pointillism, the colors are selected based on their neighbors to mix as an average illusory color in the human eye.

To use as a library, add the following to your Cargo.toml; add the palette_color feature to enable Lab color quantization.

[dependencies.rscolorq]
version = "0.1"
default-features = false

Examples

1) Mandrill

4 quantized mandrills
Top row: Original image, RGB 2 colors
Bottom row: RGB 4 colors, RGB 8 colors

rscolorq -i mandrill.jpg -o mandrill-rgb2.png -n 2 --auto -s 0 --iters 5
rscolorq -i mandrill.jpg -o mandrill-rgb4.png -n 4 --auto -s 0 --repeats 3
rscolorq -i mandrill.jpg -o mandrill-rgb8.png -n 8 --auto -s 0 --iters 5

The --iters and --repeats options can be used to increase their values over the default to improve the quality of output. --auto sets the dithering level based on the image size and desired palette size. The --seed or -s option sets the random number generator seed; otherwise, it's seeded randomly.

2) Palette swatches and fixed palette

Palette swatches can be generated by passing --op plus a filename. --width and --height can be passed to specify the width and height of the resulting palette image. The following swatches are the colors that comprise 4 and 8 color dithered images in the bottom row of the previous image.

4 color swatch
8 color swatch

rscolorq -i mandrill-resize.jpg --op mandrill-rgb4-pal.png -n 4 --auto -s 0 --repeats 3
rscolorq -i mandrill-resize.jpg --op mandrill-rgb8-pal.png -n 8 --auto -s 0 --iters 5 -p

Passing the --print or -p flag will print the hexadecimal colors to the terminal as seen in the second example above. If no --output or -o is passed, the dithered image will not be saved to a file.

b5c970,191821,b7cbe7,6d7f7b,5db7f0,4e5936,f05131,939bcc

Custom color palette

You can supply your own palette to dither with by passing --colors or -c followed by a list of hexadecimal colors as in the following example.

2 tone mandrill
Original image on the left, fixed palette on the right.

rscolorq -i scenic.jpg -o mountain-pal.png -c FFBF82,09717E --auto -s 0 --iters 5`

3) Gradients

4 quantized rainbow gradients
Top row: Original image, RGB 4 colors, RGB 8 colors.
Bottom row: Lab 4 colors, Lab 8 colors.

rscolorq -i rainbow.png -o rainbow-rgb4.png -n 4
rscolorq -i rainbow.png -o rainbow-rgb8.png -n 8 --iters 8 --repeats 2
rscolorq -i rainbow.png -o rainbow-lab4.png -n 4 --lab
rscolorq -i rainbow.png -o rainbow-lab8.png -n 8 --lab --iters 8 --repeats 2

3 greyscale gradients
Left to right: Original image, 2 colors filter size 3, 2 colors filter size 5.

Features

  • use RGB or Lab color space for calculations
  • option to dither based on fixed color palette supplied by the user
  • seedable RNG for reproducible results
  • print the palette colors to the command line in hexadecimal
  • create a palette swatch image from the dither colors

Limitations

It's "slow"

  • Larger images or images with smooth transitions/gradients will take longer. Higher palette sizes will take longer.
  • The algorithm is suited towards retaining detail with smaller color palettes. You can still use it on larger images but be aware it's not close to real-time unless the image is small.

Filter size 1x1

  • Doesn't produce an image resembling the input, nor does the original.

Filter size 5x5

  • Doesn't always converge.
  • I'm unsure if this is an error in this implementation or a problem with the random number generator being used. The original implementation may take a while but eventually completes with filter size 5.
  • Any help on this would be appreciated.

Troubleshooting

If you get an invalid color error or hex color length error with the command line tool, try enclosing the color string in quotes.

For example, instead of -c 000000,ffffff use -c '000000,ffffff'.

License

This crate is licensed under either

at your option.

Copyright of the original images is property of their respective owners.

Dependencies

~0.3–1.2MB
~23K SLoC