Find Visually Duplicate Images — Bulk Image Downloader Pro
The same product photo can appear at five different URLs: a thumbnail, a CDN resize, a WebP variant, and two re-saved JPEGs with new filenames. Filename and URL matching will not catch that. To find visually duplicate images, you need comparison by appearance, not by string.
The Perceptual Duplicate Finder on the options page Duplicate Finder tab in Bulk Image Downloader Pro does exactly that. It scores how alike two images look, groups likely copies, and lets you keep one version and remove the rest from your task lists.
Why visual matching matters
Web scrapes and URL imports often collect redundant files:
- the same hero image at multiple resolutions
- JPEG and WebP versions of one picture
- CDN paths that point to identical pixels
- re-uploaded assets with unrelated filenames
Cleaning those by hand does not scale. Visual duplicate detection finds copies your eyes would recognize, even when metadata differs.
How the finder judges similarity
Each image is analyzed into many compact fingerprints and descriptors. The scorer blends them into one similarity verdict instead of trusting a single test.
Structural hashes
Three perceptual hashes fingerprint layout and tone in ways that often survive resize and recompression:
- dHash (difference hash) from pixel gradients
- aHash (average hash) from overall brightness
- pHash from low-frequency structure
A center-focused dHash variant also helps when crops differ slightly. Hashes are fast first-pass signals, not the whole story.
Color signals
- Histogram compared with Earth Mover’s Distance for palette similarity
- Dominant colors with coverage
- Spatial color palette across a small grid
- Color moments (mean, variance, skew per channel)
- Color coherence for scattered vs clustered color
Texture, edges, and shape
- LBP (local binary patterns) for texture
- Laplacian variance for sharp vs soft focus
- Gradient histogram from edge strength
- Hu moments for shape invariants
- Edge map overlap (optional weight)
- Spatial grid of mean cell colors
Practical metadata checks
- Dimension (size and aspect ratio)
- Format (file type)
Together that is more than fifteen independent ways to ask whether two images are the same picture. A near-duplicate that fools one signal rarely fools all of them at once.
Run a scan
- open the options page and select the Duplicate Finder tab
- choose one task or All Tasks
- pick a Sensitivity preset (Balanced is the default)
- set a Keep Rule for which copy should survive in each group
- click Scan for Duplicates
Scoring runs in a Web Worker so the page stays responsive while pairs are compared.
Review groups and clean up
Results appear as duplicate sets you can sort by group size or confidence, expand or collapse, and select in bulk. Actions include:
- export or copy duplicate URLs
- download selected or non-kept images for review
- Remove Selected From Tasks or Remove Non-Kept (Use Keep Rule) to trim task URL lists
Keep rules include largest resolution (default), first detected, smallest resolution, shortest URL, and manual per group.
The UI is honest about limits: perceptual matching can occasionally miss a pair or group images you meant to keep separate. That is why sensitivity presets and weight tuning exist.
Licensing and scan scope
Duplicate Finder is a PRO options-page feature. In Basic mode the tab is visible but tier-locked. Try PRO lets you test the scanner on a capped sample (one image source in the current build). Licensed PRO scans full task batches, with a performance cap on very large lists.
Install Bulk Image Downloader Pro, run a scan on a messy task, and let visual matching surface the copies URL rules never would.
Continue reading: Find Duplicate Images With Sensitivity Settings.
Get Bulk Image Downloader Pro on the Chrome Web Store, watch the tutorial video, or visit our YouTube channel for more duplicate-finder tips.