ssimulacra2
Score how close a distorted image looks to the original
TLDR
SYNOPSIS
ssimulacra2 [original distorted]
DESCRIPTION
ssimulacra2 prints one number for how visually close a distorted image is to the original. The score runs from negative values (very strong distortion) through 100 (mathematically lossless). Both files must have the same dimensions. With no arguments the program prints usage and the meaning of the ranges.The scale, as published with the tool:- Negative: extremely low quality- 10: very low quality (obvious, annoying artifacts)- 30: low quality- 50: medium quality- 70: high quality (artifacts perceptible, not annoying)- 80: very high quality- 85: excellent quality- 90: visually lossless- 100: mathematically losslessThe metric is a perceptual comparison in XYB color space. It extends multi-scale structural similarity with two extra error maps and aggregates them with two norms. It is used to judge image codecs, including JPEG XL, against what a person would notice at 1:1.
INSTALL
CAVEATS
The two images must match in width and height. The program does not resize, crop, or align them. A score is not a file-size or PSNR figure. Different codecs can land on the same score with very different artifacts.With no arguments, ssimulacra2 explains itself and exits. It does not read a directory or a glob of frames.
HISTORY
SSIMULACRA 2 was developed at Cloudinary, with the implementation published in the ssimulacra2 repository. It is the second version of that group's subjective-similarity metric and is the basis for quality comparisons around JPEG XL.
SEE ALSO
ffmpeg(1)
