pixelmatch-fast 1.3.0

High-performance Python port of mapbox/pixelmatch for perceptual image comparison

pixelmatch-fast

High-performance Python port of mapbox/pixelmatch for perceptual image comparison

Pixelmatch is a tool that automatically highlights differences between two images while ignoring anti-aliasing artifacts.

For more information about pixelmatch capabilities and examples, see the mapbox/pixelmatch repository.

This project tries to stay up to date with the current pixelmatch version (currently matches with v7.1.0).

Similar Projects

This project is similar to pixelmatch-py. The key difference is that pixelmatch-fast is much faster by leveraging numpy for array operations and numba for JIT compilation.

Use pixelmatch-py if you want a clean port with very little dependencies, use pixelmatch-fast if you need high performance.

Installation

Install the package:

pip install pixelmatch-fast

CLI Usage

$ pixelmatch --help

Usage: pixelmatch [OPTIONS] IMG1 IMG2

  Compare two images pixel-by-pixel and visualize differences.

Options:
  --version              Show the version and exit.
  -o, --output PATH      Path to save diff image (PNG format)
  -t, --threshold FLOAT  Matching threshold (0 to 1); smaller is more
                         sensitive  [default: 0.1]
  --include-aa           Count anti-aliased pixels as different
  -a, --alpha FLOAT      Opacity of original image in diff output  [default:
                         0.1]
  --aa-color TEXT        Color of anti-aliased pixels (R,G,B)  [default:
                         255,255,0]
  --diff-color TEXT      Color of different pixels (R,G,B)  [default: 255,0,0]
  --diff-color-alt TEXT  Alternative color to differentiate between "added" and "removed" parts (R,G,B)
  --diff-mask            Draw diff over transparent background
  --help                 Show this message and exit.

Example (using test images from the mapbox/pixelmatch repository):

$ pixelmatch 1a.png 1b.png -o diff.png
Mismatched pixels: 106

The CLI exits with code 0 if images match and 1 if they differ (i.e., one or more mismatched pixels).

Library Usage

from pixelmatch import pixelmatch

# Compare two images and get mismatch count
num_diff = pixelmatch(
    "image1.png",
    "image2.png",
    output="diff.png",  # Optional: save diff image
)

print(f"Found {num_diff} mismatched pixels")

Arguments

• img1, img2 — Image paths (str or Path) or PIL Image objects to compare. Note: image dimensions must be equal.
• output — Image output for the diff. Can be a file path (str or Path) to save as PNG, a PIL Image object to fill with diff data, or None if diff output is not needed.
• threshold — Matching threshold, ranges from 0 to 1. Smaller values make the comparison more sensitive. 0.1 by default.
• includeAA — Whether to count anti-aliased pixels as different. False by default.
• alpha — Blending factor of unchanged pixels in the diff output. Ranges from 0 for pure white to 1 for original brightness. 0.1 by default.
• aa_color — Tuple of (R, G, B) color for anti-aliased pixels in diff output. (255, 255, 0) (yellow) by default.
• diff_color — Tuple of (R, G, B) color for different pixels in diff output. (255, 0, 0) (red) by default.
• diff_color_alt — Tuple of (R, G, B) for an alternative color to use for dark on light differences to differentiate between "added" and "removed" parts. If not provided, all differing pixels use diff_color.
• diff_mask — Draw the diff over a transparent background (a mask), rather than over the original image. False by default.

Development

Install with dev dependencies (pytest, ruff):

pip install -e .[dev]

Run tests:

pytest tests/

Run tests with coverage (disables numba JIT compilation):

NUMBA_DISABLE_JIT=1 pytest tests/ --cov

Check code quality:

ruff check
ruff format --check
mypy .

Install pre-commit hook (runs ruff & mypy automatically before commits):

cp hooks/pre-commit .git/hooks/pre-commit
chmod +x .git/hooks/pre-commit

The CI workflow automatically runs tests both with and without numba enabled, ensuring both the optimized and fallback code paths are tested.

License

MIT