I’m looking for a tool/software that will take an image as an example and scan my hard drive to find similar images to it, just like what Google Images does on the Web.
Must run on Linux. See Find visually similar images for a given image file (on Windows) for Windows software.
findimagedupes is the usual command-line program for that on Linux and other Unix platforms.
It looks for similarity among all files. You can filter the output to retain only information about one file if you wish, I don't think that makes a huge difference in performance (the slow part is scanning all files).
To scan PNG and JPEG files under a certain directory, and only retain images similar to SOMEIMAGE.jpg, run the following commands:
cd /path/to/directory
find . \( -name '*.jpg' -o -name '*.png' \) -print0 |
findimagedupes -0 - |
grep -F 'SOMEIMAGE.jpg'
If you're going to do that often, findimagedupes can store image fingerprints in a database, which makes searches a lot faster once the database is built. See the manual for more details.
findimagedupes works for me like a charm, here is its documentation: manpages.ubuntu.com/manpages/precise/man1/…
-a option lets you just check your database for hits against a short list of files, which will be quicker than checking them all - something like O(m*n) for m new files and n in the database, instead of O(n^2).
I recommend geeqie, you can install via package manager such as sudo apt install geeqie.
As you can see, it able to group by Pikachu's ears even though the images quite different. Note that this directory has ~3759 images and take ~5 minutes to complete.
How to use:
cd to desired directory. Alternatively you can open geeqie first and insert the path later.geeqie to open current directory images. [UPDATE]
I just wrote a python script (Not related to geeqie) to do this. This script has robust options to let you choose hash method or Manhattan norm method, saved matched/not-match output as symbolic link to specific directory, caching file(You can compare 10000+ images in seconds !), and specify acceptable percentage. Currently I only make it works in Linux.
geeqie is quite good. If only the Move to dialog box took me to the right working directory I'd find this useful.
Commented
Mar 14, 2021 at 0:36
A little DIY but you can do this sort of thing with python plus either OpenCV or Numpy - in either case the approach is the same:
Add a GUI if you desire.
NOTE: This does not work anymore. I couldn't get it to run for most files anymore, it seems like this tool is broken.
There is Libpuzzle, that will do the comparison part, but not the search for all images part. To install it use the package libpuzzle-bin, that will give you the command line tool.
It is a library and a command line tool that is designed for the purpose of comparing two images and listing their similarity. I used it to find files that were very similar (logos) a while ago, that worked well.
If you want to use it, you must use additional software that calls libpuzzle with the appropriate options.
I wrote a script that can be used for a find -exec command. The script goes like this (I have it in a file called simple-pd.sh):
#!/bin/bash
if (( $(echo "0.80<`puzzle-diff \"$1\" \"$2\"`" | bc -l) )) ; then
exit 0
fi
exit 1
it is invoked like this:
find . -regex ".*jpg\|.*gif\|.*png" -exec ./simple-pd.sh MYIMAGE {} \; -print
where you replace MYIMAGE with the image you want to have the comparison on. The script needs executable rights (chmod +x simple-pd.sh)
The tool (together with the scripts here) will do what you want most of the time.
While I do like the tool it has its downsides. It can't read all files (I had problems with jpgs sometimes) and is irritating when it fails.
The parameters for puzzle-diff are hardly documented, you will need to test on some images that you want to have and some that you don't to see which ones are the best for your task.
My Script assumes that an image that is very similar (0.80) is to be reported as YES and the rest as NO. You might want to reduce that number to fit your needs.
dupeGuru PE. Plus it's open-source. :)
dupeGuru Picture Edition (PE for short) is a tool to find duplicate pictures on your computer. dupeGuru PE is a big brother of dupeGuru. It works like dupeGuru, but is specialized for duplicate pictures matching. dupeGuru PE runs on Windows, Mac OS X and Linux.
dupeGuru PE is efficient. Not only can dupeGuru PE find exact matches, but it can also find duplicates among pictures of different kind (PNG, JPG, GIF etc..) and quality. On Mac OS X, dupeGuru PE can scan your iPhoto library.
dupeGuru PE is customizable. You can tweak its matching engine to find exactly the kind of duplicates you want to find. The Preference page of the help file lists all the scanning engine settings you can change.
dupeGuru PE is safe. Its engine has been especially designed with safety in mind. Its reference directory system as well as its grouping system prevent you from deleting pictures you didn't mean to delete.
Compare any picture format. dupeGuru PE supports JPG, PNG, TIFF, GIF and BMP formats. All these formats can be compared together. The Mac OS X version of dupeGuru PE also supports PSD and RAW (CR2 and NEF) formats.
Your iPhoto and Aperture libraries are supported.
Do whatever you want with your duplicates. Not only can you delete duplicates files dupeGuru finds, but you can also move or copy them elsewhere. There are also multiple ways to filter and sort your results to easily weed out false duplicates (for low threshold scans).
(answer is copied from https://superuser.com/questions/311633/free-visual-similarity-image-for-a-local-hard-drive-search/453251#453251 - all credit to Pat on SuperUser)
Try search by image on disk. It is browser-based (Linux, Windows, Mac etc.) program I developed. Works best with Chrome and Firefox, as other browsers might not support folder selection. You can use the program offline by downloading the web-page and hosting it at port 8080, because it does not involve server-processing of the images.
I haven't seen anyone mention Digikam, surprisingly: https://digikam.org/
It's really more of an entire photo manager, but finding similar images is one of the functionalities it offers, but there's no CLI, only GUI.
You simply "import" the folder you wish to scan for duplicates in, then choose "Find Duplicates" from the Tools menu.
