Pixel Art Maker
Generate palettes and dither any image with custom patterns.
- What is it ?
- How does it actually work ?
What is it ?
Pixel Art Maker is a two-features graphics tool designed around the style of Pixel Art. These two functionalities are :
- Generating an optimized limited color palette for an image.
- Dithering an image with an arbitrary color palette, using a user-defined pattern.
The palette generation is made using a K-means clustering algorithm that tries to find the set of K colors that average best the colors of the input image.
The dithering is a positional ordered dithering algorithm, similar to the Bayer ordered dithering algorithm. However, instead of using the optimised Bayer matrix as a pattern, any grayscale image can be used as pattern source. This is destined to artists wanting to try out some artistic patterns in their creation. Draw a pattern image, define your palette and launch the dithering. If you like the result, use it (possibly tweaking it) or try out something else !
You can build the two binaries
palette by following the
instructions in the Build section.
Assuming you are located at the root of the repo, you can generate a
16 colors palette for the example picture
paraglider.png and dither
it with an 8x8 Bayer pattern like this:
$ ./pixam-palette picture.png 16 palette.png
Now you have an optimized 16 colors palette:
You can now dither this same picture, using for example the
$ ./pixam-dither picture.png palette.png patterns/pattern-bayer8.png dithered-picture.png
The 8x8 Bayer threshold matrix looks like this (zoom 8:1) :
Although the output from the
palette command can be used as the
palette argument for the
dither command, this is not mandatory.
./pixam-palette image.png colorsCount [output.png]
image.png: path to the image file that needs an optimized palette
colorsCount: desired number of colors in the palette
output.png: name of the output palette image file (default to
./pixam-dither image.png palette.png pattern.png [output.png]
image.png: path to the image file to dither
palette.png: path to the palette to use
pattern.png: path to the image used as a dithering pattern
output.png: result image will be saved under this name (default to
How does it actually work ?
The algorithm that generate the palette is pretty straightforward as it is a strict use of the K-means clustering with pixels treated as point in the RGB space. Plus, it can be used without being understood thoroughly. Just note that the generated palette is likely to have a shorter dynamic range than the original picture, so you might want to correct this afterward with any image manipulation program.
While the palette generation does not require a precise understanding of its mechanics in order to be used, it is advised to grasp some of the underlying process when it comes to dithering. Thus the following sections.
The dithering process
The dithering algorithm works as follow :
- Retrieve every unique color from the palette image
- For each pixel in the original picture
- Find the two closest colors from the palette. These are the colors used to reproduce the original color.
- According to the brightness of the corresponding pattern pixel, choose the first or the second color.
- Fill this same pixel in the output image with the chosen color.
Step 2 needs some explanations.
Which pixel in the pattern image is associated to which pixel in the original picture ?
Well, it is the pixel with the same coordinates, modulo the width and
height of the pattern image. If the pixel coordinates in the original
(x,y), and the pattern image has a
w pixels width and
h pixels height, then the
(i,j) coordinates of the corresponding
pixel in the pattern image are such that :
i = x modulo w j = y modulo h
In other words, if you tile the original picture with the pattern image, you obtain the link between each pixel.
How does it choose between the first or second color ?
Let's take an example : the color of the original pixel is C, and the two closest colors in the palette are C1 and C2. We can calculate the distance between these colors. Let's say that C stands at 30% between C1 and C2, that is, the distance between C1 and C equals 30% of the distance between C1 and C plus the distance between C and C2.
Now it's simple, if the corresponding pattern pixel is 30% or less bright, then choose C1, else choose C2. Applying a 30% threshold to the pattern image will show as black all the pixels that will choose C1 and as white all the pixels that will choose C2.
Making a custom pattern
You can, and you are encouraged to design patterns of your own. So here are some tips that help making fun patterns :
- Avoid using plain white or plain black. A #FFFFFF white or a #000000 black will results in pixels that are forced to one of the two closest colors, thus revealing really obvious pattern in the dithering process.
- Use small images for patterns. Larger than 16x16 is not recommended...
- Each shade of gray will correspond to a unique pattern used in a unique range of color mixing. So try adjusting the different brightness percentages to fit best the ratio of drawn pixels over the total number of pixels in your pattern. For example, if you draw some 40% gray pixels, then the number of these pixels plus all the darker pixels in your pattern should make at least 40% of the pixel total.
In order to build the project, you will need the following dependencies, listed here as Debian packages:
cimg-dev: This is CImg library, the C++ image processing library used by the project.
libboost-filesystem-dev: The Boost Filesystem Library. Used to manipulate paths and files.
$ sudo apt-get install libboost-filesystem-dev cimg-dev
Then you can compile the project using the provided