About cbird

cbird is a command-line program for finding duplicate images and videos that cannot be found by general methods such as file hashing. Content-Based Image Recognition (CBIR) is used, which examines the pixels of files to get comparable features and "perceptual" hash codes.

The main features are:

• CBIR algorithms for image and video search
• Cached index in the top directory, prevents needless re-indexing
• GUI for evaluating duplicates
• Huge index/database is possible
• Comprehensive file format support
• Search inside zip files

Installing

Compile

Compile it yourself using my detailed notes.

Download

Download binaries on Github

Linux AppImage

Add execute permission and run

chmod +x cbird-0.7.0-x86_64.AppImage
./cbird-0.7.0-x86_64.AppImage -install # optional install helper
cbird [...]

• Required packages: trash-cli
• Optional packages: ocenaudio, kdenlive

AppImage Issues

• AppImage won't run (debian, Ubuntu 23.04+)
  • apt install libfuse2
• Missing libOpenGL.so.0
  • debian: apt install libopengl0
  • redhat: yum install libglvnd-opengl
• Missing window titlebar (Fedora)
  • cbird -platform wayland-egl [...]

Mac OS X 11.0+ x86

• Unzip the distribution file and run

cbird/cbird-mac [...]

Windows 10

• Unzip the distribution file and run the program
• Install helpers (optional): vlc, kdenlive

Windows PowerShell

Optional: create shortcuts for cbird

• Unzip into C:\ so you have C:\cbird\cbird.exe
• Enable script execution
  • Run PowerShell as administrator
  • Enter Set-Execution-Policy RemoteSigned
  • Close PowerShell
• Create profile script (if you don't have one)
  • Run PowerShell normally
  • New-Item -Type File $PROFILE -Force
• Add cbird shortcut to profile
  • OpenWith $PROFILE
  • Set-Alias -Name cbird -Value C:\cbird\cbird.exe
• Shortcut for pictures folder
  • function cbird-pics {cbird -use $HOME\Pictures $args}

Getting Started

Get Help

cbird -help is very detailed

Index the files in <path>, caching into <path>/_index

cbird -use <path> -update

Index files in cwd

cbird -update

MD5 Checksums

cbird -dups -show

Search cwd, default threshold

cbird -similar -show

Search cwd, lowest threshold

cbird -p.dht 1 -similar -show

Using the GUI

This is lacking documentation at the moment. But for now...

• The GUI is displayed with -show if there is a selection or results.
• GUI windows have a context menu (right click) with all available actions.
• The two deletion actions ("Delete" and "Replace") use the trash/recycler by default. There is no way to permanently delete files (not even with batch deletion commands)

Use Cases

• Find exact duplicates (using file checksums)
• Find modifications
  • General transforms: resize, rotate, crop
  • Image edits: blur, sharpen, noise, color-grade, grayscale
  • Video edits: clipping, fps change, letter boxing
• Evaluate matches
  • Compare attributes (resolution, file size, compression ratio)
  • Flip between matches
  • Zoom-in to see fine details, compression losses
  • False-color visualization of differences for fast evaluation
  • No-reference subjective quality estimate
  • Jpeg compression quality estimate
  • Align videos temporally, flip between them or play side-by-side
• File management
  • Sort/rename based on similarity
  • Rename files using regular expressions
  • Move/rename files/directories within the index

File Formats

Common formats are supported, as well as many obscure formats. The available formats will ultimately vary based on the configuration of Qt and FFmpeg.

cbird -about lists the image and video extensions. Note that video extensions are not checked against FFmpeg at runtime, so they could be unavailable.

Additionally, zip files are supported for images.

To get the most formats you will need to compile FFmpeg and Qt with the necessary options. Additional image formats are also available with kimageformats.

Link Handling

Links are ignored by default. To follow links, use the index option -i.links 1

If the search path contains links, they are only considered when scanning for changes (-update), otherwise there is no special treatment. For example, deleting a link is the same as any other deletion operation.

Duplicate inodes are not followed by default. If there are duplicate inodes in the tree, the first inode in breadth-first traversal is indexed. To follow all inodes, for example to find duplicate hard links, use -i.dups 1.

The index stores relative paths (to the indexed/root path), this makes the index stable if the parent directory changes. However, if a path contains links, or is a link itself, it is stored as-is; which may be less stable than the storing the link target. To store the resolved links instead, use i.resolve 1. This is only possible if the link target is a child of the index root.

Note that cbird does not not prevent broken links from occurring, the link check is temporary during the index update.

Using Weeds

The "weed" feature allows fast deletion of deleted files that reappear in the future. A weed record is a pair of file hashes, one is the weed/deleted file, the other is the original/retained file. When the weed shows up again, it can be deleted without inspection (-nuke-weeds)

How weeds are recorded

1. Two files are examined (matching pair) -- use -p.mm 1 or -p.eg 1 to force pairs
2. Neither file is a zip member
3. When one of the two files is deleted, it is marked as a weed of the first one

Broken weeds

There is nothing to prevent deletion of the original/retained file, so the weed record can become invalidated. If the original is no longer present, the association can be unset with the "Forget Weed" command.

cbird -weeds -show # show all weeds
cbird -nuke-weeds  # delete all weeds
cbird -similar -with isWeed true # isolate weeds in search results

Environment Variables

There are a few for power users.

• CBIRD_SETTINGS_FILE overrides the path to the settings file (cbird -about shows the default)
• CBIRD_TRASH_DIR overrides the path to trash folder, do not use the system trash bin
• CBIRD_CONSOLE_WIDTH set character width of terminal console (default auto-detect)
• CBIRD_COLOR_CONSOLE use colored output even if console says no (default auto-detect)
• CBIRD_FORCE_COLORS use colored output even if console is not detected
• CBIRD_LOG_TIMESTAMP add time delta to log messages
• CBIRD_NO_BUNDLED_PROGS do not use bundled programs like ffmpeg in the appimage/binary distribution
• QT_IMAGE_ALLOC_LIMIT_MB maximum memory allocation for image files (default 256)
• QT_SCALE_FACTOR global scale factor for UI
• TMPDIR override default directory for temporary files; used for opening zip file contents

Wish List, Bugs, Etc

Check the development notes for known bugs and feature ideas.

Report bugs or request features on github

Search Algorithms

There are several algorithms, some are better than others depending on the situation.

Discrete Cosine Transform (DCT) Hash (-p.alg dct)

Uses one 64-bit hash per image, similar to pHash. Very fast, good for rescaled images and lightly cropped images.

DCT Features -p.alg fdct

Uses DCT hashes centered on scale/rotation invariant features, up to 400 per image. Good for heavily cropped images, much faster than ORB.

Oriented Rotated Brief (ORB) Descriptors -p.alg orb

Uses 256-bit scale/rotation invariant feature descriptors, up to 400 per image. Good for rotated and cropped images, but slow.

Color Histogram -p.alg color

Uses histogram of up to 32 colors (256-byte) per image. Sometimes works when all else fails. This is the only algorithm that finds reflected images, others require -p.refl and must rehash the reflected image (very slow)

DCT Video Index -p.alg video

Uses DCT hashes of video frames. Frames are preprocessed to remove letterboxing. Can also find video thumbnails in the source video since they have the same hash type.

Template Matcher -p.tm 1

Filters results with a high resolution secondary matcher that finds the exact overlap of an image pair. This is most useful to drop poor matches from fdct and orb. Since it requires decompressing the source/destination image it is extremely slow. It can help to reduce the maximum number of matches per image with -p.mm #

How it Performs

Indexing

Indexing happens when -update is used. It can take a while the first time, however subsequent updates only consider changes.

Unused algorithms can be disabled to speed up indexing. If you have large images, you may as well enable all algorithms because image decompression dominates the process.

Table 1: Indexing 1000 6000px images, 8 GB, SSD

Arguments	Note	Time (seconds)
-update	all enabled	46
-i.algos 0 -update	md5 only	2
-i.algos 1 -update	+dct	41
-i.algos 3 -update	+dct features	44
-i.algos 7 -update	+orb features	44
-i.algos 15 -update	+color hist	46

Searching

Search speed varies with algorithm. The OpenCV search tree for ORB is quite slow compared to others. It is better suited for -similar-to to search a smaller subset suspected to have duplicates.

Table 2: Searching 1000 images

Arguments	Note	Time (milliseconds)
-similar	dct	54
-p.alg fdct -similar	dct features	200
-p.alg orb -similar	orb features	9000
-p.alg color -similar	histograms	450

Large Datasets

Indexing large sets of smaller images benefits from disabling algorithms.

Table 3: Indexing 500k 400px images in 100 zip files, 39GB, NAS

Arguments	Note	Rate (Img/s)	Time (minutes)
-i.algos 0 -update	md5 only	861	9:41
-i.algos 1 -update	+dct	683	12:11
-i.algos 3 -update	+dct features	377	22:04
-i.algos 7 -update	+orb features	348	23:56
-i.algos 15 -update	+colors	227	36:39

For N^2 search (-similar) only DCT hash is normally practical.

Table 4: Searching 500k images

Arguments	Note	Time (s)
-p.dht 1 -similar	dct, threshold 1	5.5
-p.dht 2 -similar	dct, threshold 2	5.6
-p.dht 3 -similar	dct, threshold 3	5.9
-p.dht 4 -similar	dct, threshold 4	7.1
-p.dht 5 -similar	dct, threshold 5	8.9

For K*N (K needle images, N haystack images) the slower algorithms can be practical even for large datasets. For a quick test we can select and search for the first 10 items:

cbird -p.alg fdct -select-type i -head 10 -similar-to @

Table 5: Searching for 10 images in 500k

Arguments	Note	Time (s)
-p.alg dct -p.dht 2	dct, threshold 2	1.3
-p.alg fdct -p.dht 7	dct-features, threshold 7	1.5
-p.alg orb	orb-features	84.41
-p.alg color	colors	dnf2

Release Notes

v0.7

• Mac OS X port and x86 binary package
• Use current libraries (FFmpeg, Qt6, Quazip)
• Command Line
  • describe what -filter is doing
  • cleanup some logging
• Indexing
  • Use "nice"/low priority when scanning/indexing
  • 3-4x faster jpeg indexing using idct scaling
  • Recognize more video extensions
  • Remove colon (":") workaround for powershell
  • Failsafe cache updates using write-temporary-move method
  • Fix slow cache update on network volumes
  • Fix -i.ignored to actually track all ignored stuff
• Searching
  • Add % progress for sql queries, cleanup logging
  • Search using a given hash code
  • Use numeric sort for filenames (e.g. 1.jpg sorts before 10.jpg)
  • Functions are always chainable and consistent between commands
  • Properties: add aspectRatio
  • Disable inode check on win32, too slow on network volumes
  • Fix attempt to template match video result
• Viewing
  • Add theme system, remove hardcoded theme/style
    • auto-detect dark/light theme on mac/windows
    • shade background windows when modal window is displayed
    • "-theme" to change the theme
  • Streamline context menu
  • ffplay-sbs: tons of improvements
  • Do not require index for -select-files
  • DPI-aware image scaling
  • Show creation time metadata for videos too
  • Browse the selected item in its folder
  • Use grid for folder view and fixed thumbnail aspect ratio
  • Cycle contents thumbnails when hovering in folder view
  • Add index/folder thumbnail tool
  • Fix some performance issues
  • Fix keyboard navigation in folder view
  • Fix missing match range
  • Prevent crash/hang on some invalid inputs (corrupt videos etc)
  • Fix slow qualityscore on network volumes
  • Fix potential crash on two-tone images
  • Fix potential crash/hang with corrupt video files
  • Fix elided strings display

v0.6.0

• Add weeds feature (recommended by r/user/traal)
• Search: Add max-threshold -p.mt which produces a result for every needle until threshold is exceeded
• Indexer: Use skip-loop-filter to speedup video indexing ~20%
• Indexer: Option to show all ignored files
• Indexer: Handle links sanely by default, add options i.links, -i.dups, -i.resolvefor other cases.
• Filters: Add -head/-tail
• Image view: difference image can align to template match
• Video view: fix issues with unequal frame rates
• Video view: export the aligned pair to Kdenlive
• Speedups: video compare view, differences image, -group-by

Footnotes

1. OpenCV search tree only partially cached on disk, slow to start ↩

2. Color search lacks a search tree, not suitable for large sets ↩