Color - quantization dithering - add algorithms, unify options

[user0]

XnViewMP 1.7.1
Windows11 (23H2), display resolution 3840x2160, dpi scale 150%

Color depth - quantization and dithering
Dithering options are different for Grayscale and Indexed colors in 'Change color depth...' and 'Export...' dialogs.
It is also not indicated which quantization algorithm is used, even though their quality/speed can differ significantly.


Suggestion

color_depth_add_quantization_dithering.png

Color Quantization

• add indication of which algorithm is in use OR add selector if more than one is available

• add more quantization algorithms
  eg NeuQuant, Wu (like in RIOT)
  
  Algorithms overview

  Algorithm	Quality	Speed	Typical use today	Notes
  libimagequant / pngquant	Excellent	Medium	PNG/GIF export	Widely considered the best practical modern quantizer
  K-means (optimized/perceptual)	Excellent	Slow–Medium	High-quality offline processing	Often produces the best palettes when carefully tuned
  Wu quantization	Very good	Very fast	Real-time apps	Excellent speed/quality balance
  NeuQuant	Very good	Medium	GIF/video palettes	Neural-network SOM approach
  Median Cut / MMCQ	Good	Fast	UI palettes, thumbnails	Simple and still widely used
  Octree	Fair–Good	Very fast	Embedded/realtime	Memory-efficient
  PNN / Fast Pairwise Nearest Neighbor	Excellent	Slow	Research/high quality	Can outperform classic algorithms visually

Dithering

• unify Dithering options for Grayscale and Indexed colors in:
  - Change color depth...
  - Export...

• add more Dithering algorithms
  eg
  - Blue noise [1, 2]
  - Atkinson
  - Stucki
  - Jarvis

• (optional) add standalone 'Dither...' action in Effects/Batch convert Actions (like in GIMP)
  to have all dither algos available in the app in a single list AND apply to an existing 8-bit image w/o additional palette re-quantization.
  Note: a caveat in GIMP is that it accepts full-color images as well, but does not reveal which color quantization is used (in theory user may have preferences).
  
  
  
  Comparison

  Rank	Algorithm	Type	Strengths	Weaknesses	Best Use Cases
  1	Blue-noise hybrid	Modern / Hybrid	Best visual quality, minimal patterns, excellent gradients, GPU-friendly	More complex implementation	Modern image viewers, HDR, animation, GPU pipelines
  2	Sierra Lite	Error diffusion	Excellent quality/speed balance, fewer artifacts than Floyd	Slightly softer detail	Realtime image viewers, general-purpose dithering
  3	Stucki	Error diffusion	Very smooth gradients, high photographic quality	Slower than Floyd/Sierra Lite	High-quality photo quantization
  4	Jarvis–Judice–Ninke (JJN)	Error diffusion	Very smooth gradients, low visible artifacts	Higher CPU cost	Offline processing, high-quality still images
  5	Floyd–Steinberg	Error diffusion	Fast, simple, widely supported	Directional “worm” artifacts	Classic default dithering
  6	Atkinson	Error diffusion	Sharp retro look, preserves edges	Reduced tonal accuracy	Retro graphics, monochrome, pixel art
  7	Bayer ordered dithering	Ordered	Extremely fast, deterministic, GPU-friendly	Visible repeating patterns	Realtime rendering, UI, shaders, video

ref
https://en.wikipedia.org/wiki/Color_quantization
https://en.wikipedia.org/wiki/Dither
https://surma.dev/things/ditherpunk/
https://blog.maximeheckel.com/posts/the ... ading-web/





Example

• color depth - quantization, dithering
  other_apps_color_depth.png