10 releases

0.2.6 Apr 3, 2024
0.2.5 Nov 29, 2023
0.1.2 Nov 27, 2023

#1878 in Command line utilities

38 downloads per month

GPL-3.0-or-later

28KB
593 lines

Crates.io AUR version

simagef is a CLI tool for finding similar images. It takes a list of image paths and returns either groups or pairs of images that are similar.

It uses the image-match crate to generate signatures for images and compare them. An option to match images using pixel comparisons via the image-compare crate is also available.

simagef is faster than the findimagedupes Perl script, but might not produce identical results - not even across the signature/pixel modes, and is not designed to be a drop-in replacement.

Performance

With default settings on a laptop with an Intel Core i7-11800-H processor, simagef can compare 12,399 small-to-medium sized images in 47 seconds, for a total of 76,861,401 comparisons.

Installation

Install via cargo on most platforms:

cargo install simagef

Install on Arch Linux:

yay simagef-bin

Usage

To compare a set of images:

simagef a.png b.png c.png

You can set the threshold of similarity with the -t or --threshold options:

simagef -t 50 a.png b.png c.png

You can additionally read filenames from stdin if - appears in the filenames list. For example using it with the fd command for finding files:

fd . ~/my_images | simages base.png base2.png -

You can use the (slower) pixel-based algorithm with the -m or --pixels flag.

If you want only the pairs of images without the groupings, use the -p or --pairs flag.

You can specify an external image viewer for comparing groups of images using -e or --exec. You can also provide command line arguments:

simagef -e "gwenview -s" ~/Pictures/*

This will launch the executable with the groups (or pairs) of filenames as arguments. It will launch the executable again for the next group once the previous executable exits.

Caveats

  • The crossbeam channels used in the code are unbounded, which I consider to be a bug.

  • The groups are created using a recursive graph algorithm.

Dependencies

~8.5MB
~135K SLoC