Crate butteraugli_oxide 

Butteraugli

Butteraugli is a perceptual image quality metric developed by Google. This is a Rust port of the butteraugli algorithm from libjxl.

The metric is based on:

• Opsin: dynamics of photosensitive chemicals in the retina
• XYB: hybrid opponent/trichromatic color space
• Visual masking: how features hide other features
• Multi-scale analysis: UHF, HF, MF, LF frequency components

Quality Thresholds

• Score < 1.0: Images are perceived as identical
• Score 1.0-2.0: Subtle differences may be noticeable
• Score > 2.0: Visible difference between images

Example

use butteraugli_oxide::{compute_butteraugli, ButteraugliParams};

// Create two 8x8 RGB images (must be 8x8 minimum)
let width = 8;
let height = 8;
let rgb1: Vec<u8> = (0..width * height * 3).map(|i| (i % 256) as u8).collect();
let rgb2 = rgb1.clone(); // Identical images

let params = ButteraugliParams::default();
let result = compute_butteraugli(&rgb1, &rgb2, width, height, &params).unwrap();

// Identical images should have score ~0
assert!(result.score < 0.01);

Comparing Different Images

use butteraugli_oxide::{compute_butteraugli, ButteraugliParams, BUTTERAUGLI_GOOD, BUTTERAUGLI_BAD};

let width = 16;
let height = 16;

// Original image - gradient
let original: Vec<u8> = (0..width * height)
    .flat_map(|i| {
        let x = i % width;
        [(x * 16) as u8, 128, 128]
    })
    .collect();

// Distorted image - add noise
let distorted: Vec<u8> = original.iter()
    .map(|&v| v.saturating_add(10))
    .collect();

let result = compute_butteraugli(&original, &distorted, width, height, &ButteraugliParams::default())
    .expect("valid image data");

if result.score < BUTTERAUGLI_GOOD {
    println!("Images appear identical to humans");
} else if result.score > BUTTERAUGLI_BAD {
    println!("Visible difference detected");
}

References

• https://github.com/google/butteraugli
• https://github.com/libjxl/libjxl

Re-exports

pub use crate::image::Image3F;

pub use crate::image::ImageF;

Modules

blur

Gaussian blur implementation for butteraugli.

consts

Constants for butteraugli and XYB color space.

image

Image buffer types for butteraugli.

malta

Malta edge-aware filter for butteraugli.

mask

Visual masking functions for butteraugli.

opsin

Butteraugli OpsinDynamicsImage implementation.

psycho

Multi-scale psychovisual decomposition for butteraugli.

xyb

XYB color space conversion for butteraugli.

Structs

ButteraugliParams

Butteraugli comparison parameters.

ButteraugliResult

Butteraugli image comparison result.

Enums

ButteraugliError

Error type for butteraugli operations.

Constants

BUTTERAUGLI_BAD

Quality threshold for “bad” (visible difference).

BUTTERAUGLI_GOOD

Quality threshold for “good” (images look the same).

Functions

butteraugli_fuzzy_class

Converts butteraugli score to fuzzy class value.

compute_butteraugli

Computes butteraugli score between two sRGB images.

compute_butteraugli_linear

Computes butteraugli score between two linear RGB images.

score_to_quality

Converts butteraugli score to quality percentage (0-100).

srgb_to_linear

Converts sRGB u8 value to linear RGB f32.