Image Deduplicator

Windows desktop application (GTK 4) for finding and removing visually similar images.

Near-duplicates are clustered by perceptual hash or ORB feature matching, then reviewed in a card gallery.

Русский

Features

• Comparison modes: Average hash (aHash), DCT pHash, ORB (OpenCV)
• Configurable similarity threshold and worker thread count
• Union-Find clustering of similar pairs
• Persistent caches (hash_cache.bin, feature_cache.bin)
• Virtualized results gallery with hover preview
• Group navigation: Prev/Next; Shift ±10; Ctrl+Shift ±100
• Recursive scan, exclude directories, Unicode paths
• UI languages: English, Russian; Graphite light/dark themes

Supported formats: JPEG, PNG, BMP.

Requirements

• MSYS2 MinGW64
• Packages: mingw-w64-x86_64-gtk4, mingw-w64-x86_64-libpng, mingw-w64-x86_64-libjpeg-turbo, mingw-w64-x86_64-opencv, mingw-w64-x86_64-pkgconf
• Graphite GTK theme as Graphite-Light / Graphite-Dark under C:\msys64\mingw64\share\themes

Build

build.bat

Produces a portable tree in bin\:

• Image_Deduplicator.exe — launcher (adds DLL\ to PATH, then starts the app)
• Image_Deduplicator_app.exe — application binary
• DLL\ — MinGW/GTK/OpenCV runtime libraries
• share\, lib\ — schemas, themes, icons, gdk-pixbuf loaders

bin\Image_Deduplicator.exe

Usage

1. Select a directory (optionally exclude subfolders).
2. Configure threshold, algorithm, and threads in settings.
3. Start the scan.
4. Review groups, select unwanted images, delete or open.

Algorithms

Algorithm	Use case	Typical threshold
Average (aHash)	Resize, JPEG recompression	85–95%
DCT (pHash)	Same; often more stable than aHash	85–95%
ORB	Crops, rotations, partial overlap	20–50%

ORB pipeline: ORB descriptors → Hamming match (Lowe ratio) → RANSAC homography → score in [0, 1].

Configuration

settings.ini next to the executable:

Key	Values	Description
recursive	true / false	Scan subdirectories
hash_threshold	0–100	Minimum similarity to link images
hash_algorithm	average / dct / orb	Comparison mode
thread_count	integer	Worker threads (capped by CPU count)
app_theme	light / dark	Theme
language	en / ru	UI language
debug	true / false	Right-click thumbnail → save hash/keypoints under DEBUG\

Layout

include/          Headers
src/              Sources
build.bat         Build and package
copy_deps.ps1     Runtime DLL copy

Contributing

Issues and pull requests: GitHub.

License

MIT. See LICENSE.