findimagedupes - Finds visually similar or duplicate images


findimagedupes [option ...] [--] [ - | [file ...] ]

      -f, --fingerprints=FILE    -c, --collection=FILE
      -M, --merge=FILE           -p, --program=PROGRAM
      -P, --prune                -s, --script=FILE
      -a, --add                  -i, --include=TEXT
      -r, --rescan               -I, --include-file=FILE
      -n, --no-compare
                                 -q, --quiet
      -t, --threshold=AMOUNT     -v, --verbosity=LIST

      -0, --null                 -h, --help
      -R, --recurse                  --man

With no options, compares the specified files and does not use nor update any fingerprint database.

Directories of images may be specified instead of individual files; Sub-directories of these are not searched unless --recurse is used.


If you use linux, your distribution may include a prepackaged version. For example, Debian and Ubuntu do.

Otherwise, at a minimum you'll need Perl with the modules listed at the top of the findimagedupes script. Also the GraphicksMagick package.

You may need to change Inline's DIRECTORY to point somewhere else. Read the Inline module documentation for details.


-0, --null

If a file - is given, a list of files is read from stdin.

Without -0, the list is specified one file per line, such as produced by find(1) with its -print option.

With -0, the list is expected to be null-delimited, such as produced by find(1) with its -print0 option.

-a, --add

Only look for duplicates of files specified on the commandline.

Matches are also sought in any fingerprint databases specified.

-c, --collection=FILE

Create GQView collection FILE.gqv of duplicates.

The program attempts to produce well-formed collections. In particular, it will print a warning and exclude any file whose name contains newline or doublequote. (In this situation, gqview(1) seems to create a .gqv collection file that it silently fails to read back in properly.)

-d, --debug=OPTS

Enable debugging output. Options OPTS are subject to change. See the program source for details.

-f, --fingerprints=FILE

Use FILE as fingerprint database.

May be abbreviated as --fp or --db.

This option may be given multiple times when --merge is used. (Note: FILE could contain commas, so multiple databases may not be specified as a single comma-delimited list.)

-h, --help

Print usage and option sections of this manual, then exit.

-i, --include=TEXT

TEXT is Bourne-shell code to customise --script.

It is executed after any code included using --include-file.

May be given multiple times. Code will be concatenated.

-I, --include-file=FILE

FILE is a file containing Bourne-shell code to customise --script.

It is executed before any code included using --include.


Display the full documentation, using default pager, then exit.

-M, --merge=FILE

Takes any databases specified with --fingerprints and merges them into a new database called FILE. Conflicting fingerprints for an image will cause one of two actions to occur:

  1. If the image does not exist, then the entry is elided.

  2. If the image does exist, then the old information is ignored and a new fingerprint is generated from scratch.

By default, image existence is not checked unless there is a conflict. To force removal of defunct data, use --prune as well.

A list of image files is not required if this option is used. However, if a list is provided, fingerprint data for the files will be copied or (re)generated as appropriate.

When --merge is used, the original fingerprint databases are not modified, even if --prune is used.

If multiple fingerprint databases are to be used but the merge output is not required, specify: --merge=/dev/null

See also: --rescan

-n, --no-compare

Don't look for duplicates.

-p, --program=PROGRAM

Launch PROGRAM (in foreground) to view each set of dupes.

PROGRAM must be the full path to an existing executable file. For more flexibility, see the --include and --include-file options.

See also: --script

-P, --prune

Remove fingerprint data for images that do not exist any more. Has no effect unless --fingerprints or --merge is also used.

Databases specified by --fingerprints are only modified if --merge is not used.

-q, --quiet

This option may be given multiple times.

Usually, progress, warning and error messages are printed on stderr. If this option is given, warnings are not displayed. If it is given twice or more, errors are not displayed either.

Information requested with --verbosity is still displayed.

-R, --recurse

Use --recurse to search recursively for images inside subdirectories. For historical reasons, the default is to not do so. To avoid looping, symbolic links to directories are never followed.

-r, --rescan

(Re)generate all fingerprints, not just any that are unknown.

If used with --add, only the fingerprints of files specified on the commandline are (re)generated.

Implies --prune.

-s, --script=FILE

When used with --program, PROGRAM is not launched immediately. Instead sh(1)-style commands are saved to FILE. This script may be edited (if desired) and then executed manually.

When used without --program, two skeletal shell functions are generated: VIEW simply echo(1)s its arguments; the empty function END runs after files-processing is finished.

To display to terminal (or feed into a pipe), use - as FILE.

If --script is not given, the script is still created in memory and is executed immediately. So, with the default VIEW and END functions, lines containing sets of duplicates are displayed. See: EXAMPLES

See also: --include, --include-file

-t, --threshold=AMOUNT

Use AMOUNT as threshold of similarity. Append % to give a percentage or b for bits. For backwards compatibility, a number with no unit is treated as a percentage. Percentage is the minimum required for a match; bits is the maximum that may differ: bits=floor(2.56(100-percent))

A fractional part may be given but it is only accurate to 100/256 (0.390625) for percentage and it is meaningless for bits. Default is 90% (25b) if not specified.

-v, --verbosity=LIST

Enable display of informational messages to stdout, where LIST is a comma-delimited list of:


Display the checksum for each file, as per md5sum(1).

fingerprint | fp

Display the base64-encoded fingerprint of each file.

Alternatively, --verbosity may be given multiple times, and accumulates. Note that this may not be sensible. For example, to be useful, md5 output probably should not be merged with fingerprint data.


Display the program version, then exit.


findimagedupes compares a list of files for visual similarity.

To calculate an image fingerprint:
 1) Read image.
 2) Resample to 160x160 to standardize size.
 3) Grayscale by reducing saturation.
 4) Blur a lot to get rid of noise.
 5) Normalize to spread out intensity as much as possible.
 6) Equalize to make image as contrasty as possible.
 7) Resample again down to 16x16.
 8) Reduce to 1bpp.
 9) The fingerprint is this raw image data.
To compare two images for similarity:
 1) Take fingerprint pairs and xor them.
 2) Compute the percentage of 1 bits in the result.
 3) If percentage exceeds threshold, declare files to be similar.





Usage information was requested (--help or --man), or there were warnings.


Invalid options or arguments were provided.


Runtime error.

Any other return values indicate an internal error of some sort.


To be written.


findimagedupes -R -- .

Look for and compare images in all subdirectories of the current directory.

find . -type f -print0 | findimagedupes -0 -- -

Same as above.

findimagedupes -i 'echo "# sort: manual"' -i 'VIEW(){ for f in "$@";do echo \"file://$f\";done }' -- *.jpg > dupes.gqv

Use script hooks to produce collection-style output suitable for use with gthumb(1).


To be written.


There is a memory leak somewhere.

Killing the program may corrupt the fingerprint database(s).

The program does not lock the fingerprint database although concurrent write access to it is unsafe.

GraphicsMagick does not expose its auto-orient functionality to Perl.

Changing version of GraphicsMagick invalidates fingerprint databases.


Directory recursion is deliberately not implemented: Composing a file-list and using it with - is a more flexible approach.

Repetitions are culled before comparisons take place, so a commandline like findimagedupes a.jpg a.jpg will not produce a match.

The program needs a lot of memory. Probably not an issue, unless your machine has less than 128MB of free RAM and you try to compare more than a hundred-thousand files at once (and the program will run quite slowly with that many files anyway---about eight hours initially to generate fingerprints and another ten minutes to do the actual comparing).

Fingerprinting images is a bottleneck but unfortunately the program was not written with parallel processing in mind. For a workaround, see:


find(1), md5sum(1)

gqview - GTK based multiformat image viewer

gthumb - an image viewer and browser for GNOME


Jonathan H N Chin <>


 Copyright © 2006-2022 by Jonathan H N Chin <>.

 This program is free software; you may redistribute it and/or modify
 it under the terms of the GNU General Public License as published by
 the Free Software Foundation; either version 3 of the License, or
 (at your option) any later version.

 This program is distributed in the hope that it will be useful,
 but WITHOUT ANY WARRANTY; without even the implied warranty of
 GNU General Public License for more details.

 You should have received a copy of the GNU General Public License
 along with this program; if not, write to the Free Software
 Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA


This code has been written from scratch. However it owes its existence to findimagedupes by Rob Kudla and uses the same duplicate-detection algorithm.