ssimulacra2 0.2.0

Python implementation of SSIMULACRA2 image quality metric

SSIMULACRA2

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

Description

SSIMULACRA2 is a full-reference image quality metric that aims to estimate human perception of image quality, specifically focusing on compression artifacts. This Python package provides a clean, efficient implementation that closely follows the original C++ algorithm.

SSIMULACRA2 scores range from 100 (perfect quality) down to negative values (extremely poor quality):

• negative scores: extremely low quality, very strong distortion
• 10: very low quality (average output of cjxl -d 14 / -q 12 or libjpeg-turbo quality 14, 4:2:0)
• 30: low quality (average output of cjxl -d 9 / -q 20 or libjpeg-turbo quality 20, 4:2:0)
• 50: medium quality (average output of cjxl -d 5 / -q 45 or libjpeg-turbo quality 35, 4:2:0)
• 70: high quality (hard to notice artifacts without comparison to the original)
• 80: very high quality (impossible to distinguish from the original in a side-by-side comparison at 1:1)
• 85: excellent quality (impossible to distinguish from the original in a flip test at 1:1)
• 90: visually lossless (impossible to distinguish from the original in a flicker test at 1:1)
• 100: mathematically lossless

Installation

pip install ssimulacra2

Usage

Command Line

# Basic usage
ssimulacra2 original.png compressed.png

# For just the score without interpretation
ssimulacra2 original.png compressed.png --quiet

Python

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}")

Features

• Accurate: Closely follows the original C++ implementation
• Alpha Support: Special handling for images with transparency
• Multi-scale Analysis: Examines image quality at multiple resolution scales
• XYB Color Space: Uses a perceptual color space for more accurate results
• Easy to Use: Simple command-line interface and Python API

Requirements

• Python 3.8+
• NumPy
• SciPy
• Pillow (PIL)
• scikit-image
• OpenCV

Performances

Size : 1024x768

v0.2.0

Time (mean ± σ):     671.5 ms ±   8.2 ms    [User: 566.1 ms, System: 100.2 ms]
Range (min … max):   659.6 ms … 683.4 ms    10 runs

v0.1.0

Time (mean ± σ):     22.447 s ±  0.197 s    [User: 22.263 s, System: 0.069 s]
Range (min … max):   22.186 s … 22.723 s    10 runs 

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.