go-quantise

Go Quantise

Overview

Library to quantise images using algorithms from Virmajoki, O., & Franti, P. (2003). Multilevel thresholding by fast PNN-based algorithm. Image Processing: Algorithms and Systems II. doi:10.1117/12.477741

Install

go get github.com/fiwippi/go-quantise/pkg/quantisers
go mod tidy

Usage

Errors ignored for brevity

// Read in an image
reader, _ := os.Open("fish.jpg")
img, _, _ := image.Decode(reader)

// Quantise the image with PNN
colours := pnn.QuantiseColour(img, 10)
quantisedImg, _ := quantisers.ImageFromPalette(img, colours, quantisers.NoDither)
paletteImg := quantisers.ColourPaletteImage(colours, 200)

// Save the recreated image
savedImg, _ := os.Create("fish-quantised.jpg")
png.Encode(savedImg, quantisedImg)

// Save the colour palette image
savedImg, _ = os.Create("fish-palette.jpg")
png.Encode(savedImg, paletteImg)

Result

Input

Output

Raw

Dithered (Floyd-Steinberg)

Palette

Notes

The quantisation algorithms implemented from the paper are:

• FastOtsu
• Lloyd Max Quantiser (LMQ)
• PNN (In RGB and LAB space)

Available dithering algorithms are:

• Floyd-Steinberg
• Floyd-Steinberg Serpentine
• Bayer 2x2 Matrix
• Bayer 4x4 Matrix
• Bayer 8x8 Matrix

Due to limitations of each algorithm:

• Otsu only supports greyscale quantisation with m = 1
• LMQ only supports greyscale quantisation
• Images with m = 1 do not dither.

Sections of this code are adapted from Miller Chan's code found here. A big thank you to him for his help in explaining sections of it.

License

MIT