Crate iqa 

iqa provides a single, ergonomic API over the patchwork of visual quality assessment metrics available in the Rust ecosystem.

Every metric consumes the same Image<F> type, parameterized by a PixelFormat marker. The format fixes an image’s color space, channel layout, and bit depth at the type level, and a metric requires its two inputs to share one format. Comparing images that differ in any of those properties — a classic source of meaningless scores — is therefore a compile error, not a silent surprise. Only a dimension mismatch, which cannot be known at compile time, remains a runtime error.

Decoding external image formats into an Image is out of scope: callers build one directly from a sample buffer via Image::srgb8 and friends.

Features

Each metric is gated behind a Cargo feature:

• psnr (default) — peak signal-to-noise ratio, a native implementation.
• ssim (default) — structural similarity index (SSIM), a native implementation.
• dssim (default) — structural dissimilarity, (1 - SSIM) / 2, a native implementation layered on ssim (which it enables).
• ms-ssim (default) — multi-scale SSIM (MS-SSIM), a native implementation that evaluates ssim over an image pyramid (and so enables ssim).
• iw-ssim (default) — information content weighted SSIM (IW-SSIM, Wang & Li 2011), a native Laplacian-pyramid extension of MS-SSIM that pools ssim’s contrast-structure maps by a Gaussian-Scale-Mixture information weight (and so enables ssim).
• psnr-hvs-m (default) — PSNR-HVS-M (Ponomarenko et al. 2007), a native DCT-domain PSNR with a contrast-sensitivity function and contrast masking.
• ciede2000 (default) — CIEDE2000 (ΔE₀₀), a native implementation of the CIE’s perceptual color difference: per-pixel sRGB→CIELAB (D65), mean-pooled.
• ssimulacra2 (default) — SSIMULACRA2, bound via FFI to the vendored C++ reference. Enabling it requires the third_party/ git submodules and a C++ toolchain; see the project README.
• butteraugli (default) — Butteraugli, bound via FFI to libjxl’s vendored C++. Shares the same native build as ssimulacra2.

Structs

ButteraugliOptions

Tuning parameters for butteraugli.

Ciede2000Options

Options controlling a CIEDE2000 computation.

DssimOptions

Options controlling a DSSIM computation.

Gray8

8-bit grayscale, one channel per pixel.

Gray16

16-bit grayscale, one channel per pixel.

Image

A decoded image of pixel format F.

IwssimOptions

Options controlling an IW-SSIM computation.

MsssimOptions

Options controlling an MS-SSIM computation.

PsnrHvsOptions

Options controlling a PSNR-HVS-M computation.

PsnrOptions

Options controlling a PSNR computation.

Rgba8

8-bit sRGB with an alpha channel, four channels per pixel (R, G, B, A).

Rgba16

16-bit sRGB with an alpha channel, four channels per pixel (R, G, B, A).

Srgb8

8-bit sRGB, three channels per pixel (R, G, B).

Srgb16

16-bit sRGB, three channels per pixel (R, G, B).

SsimOptions

Options controlling an SSIM computation.

Enums

BitDepth

Bits per sample of an Image.

Channels

Channel layout of an Image.

ColorSpace

Color space the pixel values are encoded in.

Error

Errors produced when constructing images or computing metrics.

PsnrHvsMode

Which signal PSNR-HVS-M is computed over.

PsnrMode

Which signal PSNR is computed over.

SsimMode

Which signal SSIM is computed over.

Traits

ButteraugliInput

A pixel format that butteraugli can score.

PixelFormat

A compile-time description of how an Image’s pixels are laid out and interpreted.

Sample

A raw image sample type: either u8 or u16.

Ssimulacra2Input

A pixel format that ssimulacra2 can score.

Functions

butteraugli

Computes the Butteraugli distance between reference and distorted.

ciede2000

Computes the mean CIEDE2000 (ΔE₀₀) color difference between reference and distorted.

dssim

Computes the structural dissimilarity (DSSIM) between reference and distorted as (1 - SSIM) / 2.

iwssim

Computes the IW-SSIM between reference and distorted.

msssim

Computes the MS-SSIM between reference and distorted.

psnr

Computes the PSNR (in decibels) between reference and distorted.

psnr_hvs_m

Computes the PSNR-HVS-M (in decibels) between reference and distorted.

ssim

Computes the mean SSIM between reference and distorted.

ssimulacra2

Computes the SSIMULACRA2 score between reference and distorted.

Type Aliases

Result

Convenience alias for results returned by this crate.