SSIMULACRA2

A Python implementation of SSIMULACRA2 (Structural SIMilarity Unveiling Local And Compression Related Artifacts) - a perceptual image quality metric designed to detect and measure compression artifacts.

Overview

SSIMULACRA2 is a full-reference image quality metric that mimics human perception of image quality, focusing specifically on compression artifacts. This Python package provides an efficient implementation that closely follows the original C++ algorithm from the JPEG XL project.

For detailed information about the metric and score interpretation, please refer to the original SSIMULACRA2 implementation in the JPEG XL repository.

Installation

pip install ssimulacra2

Usage

Command Line

# Basic usage (outputs only the score)
ssymulacra2 original.png compressed.png

# With detailed quality interpretation
ssymulacra2 original.png compressed.png --verbose

# Show version information
ssymulacra2 --version

Note: The command name is ssymulacra2 (with a 'y') to avoid conflict with the C++ implementation named ssimulacra2.

Python API

from ssimulacra2 import compute_ssimulacra2, compute_ssimulacra2_with_alpha

# Basic usage
score = compute_ssimulacra2("original.png", "compressed.png")
print(f"Quality score: {score:.2f}")

# For images with alpha channel (automatically uses both dark and light backgrounds)
score = compute_ssimulacra2_with_alpha("original.png", "compressed.png")
print(f"Quality score with alpha: {score:.2f}")

Performance and Benchmarking

Performance Metrics

This implementation is optimized for speed while maintaining accuracy. Performance benchmarks for a 1024x768 image:

The dramatic speed improvement from v0.1.0 to v0.2.0 comes from better leveraging NumPy's vectorized operations.

Running Your Own Benchmarks

The package includes a performance benchmarking script that allows you to evaluate the speed of different versions:

# Benchmark current version
./performance_benchmark.sh original.png compressed.png

# Benchmark a specific version
./performance_benchmark.sh original.png compressed.png --tag v0.2.0

# Compare all available versions
./performance_benchmark.sh original.png compressed.png --full

# Customize benchmark parameters
./performance_benchmark.sh original.png compressed.png --warmup 5 --runs 20

Benchmark Script Options

Usage: ./performance_benchmark.sh <original_image> <compressed_image> [options]
Options:
  --full                Test all tagged versions simultaneously
  --tag <tag_version>   Test a specific tagged version (e.g., v0.1.0)
  --warmup <count>      Number of warmup runs (default: 2)
  --runs <count>        Number of benchmark runs (default: 10)
  --debug               Display debugging information
  --help                Show this help message

Using with press_it Benchmarks

SSIMULACRA2 is used by the press_it package's benchmark tool to evaluate image compression quality across different formats. When running press-benchmark, this Python implementation is used to calculate quality scores for compressed images (alongside C++ and Rust implementations when available).

For more comprehensive compression benchmarking, refer to the press_it documentation.

Requirements

• Python 3.8+
• NumPy
• SciPy
• Pillow (PIL)

License

This project is licensed under the BSD 3-Clause License - see the LICENSE file for details.

Acknowledgements

This implementation is based on the original SSIMULACRA2 algorithm developed for the JPEG XL project.