Dithering Algorithms Explained

The most widely used dithering algorithm, Floyd-Steinberg distributes quantization error to neighboring pixels in a 4-pixel pattern. It produces natural, balanced results ideal for photographs and detailed images. This algorithm preserves gradients well and creates organic patterns without visible structure. Best for: Photographs, portraits, realistic images, general-purpose dithering.

Developed by Apple engineer Bill Atkinson for the original Macintosh, this algorithm distributes only 6/8 of the quantization error, creating lighter, airier patterns. It produces a distinctive retro aesthetic associated with early Mac graphics and HyperCard. Best for: Vintage computing aesthetics, retro artwork, high-key images, illustrations, nostalgic effects.

Uses an 8×8 repeating threshold matrix to create distinctive geometric patterns. This algorithm produces a regular halftone-like appearance reminiscent of newspaper printing and comic books. The patterns are immediately recognizable and highly stylized. Best for: Poster art, comic book effects, retro printing aesthetics, geometric patterns, pop art.

A family of error diffusion algorithms that distribute quantization error across wider patterns than Floyd-Steinberg. Sierra uses 10 pixels across 3 rows, while Two-Row Sierra uses 7 pixels across 2 rows. These create slightly different textures with varying degrees of detail preservation. Best for: Detailed images requiring fine texture, landscapes, complex scenes.

Similar to Sierra but with different error distribution weights. Burkes emphasizes forward error propagation, creating distinctive diagonal patterns in smooth gradients. It produces a unique aesthetic between Floyd-Steinberg and Sierra. Best for: Experimental effects, unique textures, gradient-heavy images.

One of the widest error diffusion algorithms, distributing error to 12 neighboring pixels across 3 rows. This creates very smooth transitions and excellent gradient reproduction at the cost of slightly less detail preservation. Best for: Images with large smooth areas, skies, gradient backgrounds, subtle effects.

Unlike ordered dithering, which uses a fixed repeating matrix, Blue Noise uses a pre-generated Void-and-Cluster noise tile with energy concentrated at high frequencies. The result is a random-looking, grain-like pattern with no visible structure or banding — closer to analog film grain than a computer pattern. Best for: Natural, organic textures; avoiding visible halftone grids; photographic realism with a retro feel.

A high-quality error diffusion algorithm that distributes quantization error across 12 neighboring pixels in three rows using carefully tuned weights. Similar to Jarvis-Judice-Ninke but with slightly different coefficients, producing very smooth gradients with excellent detail retention and minimal noise. Best for: High-fidelity photographic dithering, smooth gradients, subtle tonal work.

A lightweight, two-pixel variant of the Sierra algorithm. It distributes error to only two forward neighbors rather than Sierra's ten, making it significantly faster while still producing pleasing results. The reduced error spread creates slightly sharper, higher-contrast output than full Sierra. Best for: Quick results, performance-sensitive use cases, high-contrast line art, and images where a sharper dither texture is desired.

A hybrid algorithm combining error diffusion with circular dot patterns. It creates an authentic bitmap halftone printing aesthetic with visible dots, similar to traditional newspaper or magazine printing. Best for: Halftone printing effects, newspaper aesthetics, comic book artwork, retro magazine looks.