Free Duplicate Image Finder — Perceptual Hash, Browser-Based
Phone photo libraries quietly accumulate duplicates: the same shot saved twice, the WhatsApp copy downloaded separately, the 'final-v2' export of the 'final' version. PikDraw's Duplicate Image Finder uses perceptual hashing (aHash) to surface visual duplicates — even when filenames differ and bytes don't match — entirely in your browser, free, with no upload.
What is the Duplicate Image Finder — Perceptual Hash, Browser-Based?
Duplicate Finder is a client-side bulk duplicate detector. Upload any number of images; the tool computes a 256-bit perceptual fingerprint per image and groups files whose fingerprints differ by less than the configured threshold. Groups are displayed side-by-side for visual review.
Key features
- Average hash (aHash) on a 16×16 grayscale grid
- Adjustable Hamming-distance threshold (0–40)
- Catches resized and recompressed copies
- Visual side-by-side group review
- Per-session — no library stored anywhere
- Unlimited file count (RAM permitting)
- 100% client-side — no upload
- Free, no signup, no watermark
How it works
Every uploaded image is decoded into a 16×16 grayscale Canvas. The mean pixel brightness is computed; each of the 256 pixels is encoded as 1 (above mean) or 0 (below mean). The resulting 256-character bit string is the image's aHash fingerprint. With all hashes computed, the tool walks the list and groups any pair whose Hamming distance is at or below the threshold. Groups of two or more are surfaced; singletons are hidden.
Why use this tool
Desktop dedupe tools (VisiPics, dupeGuru) are powerful but Windows/Mac-only. Online dedupers usually require upload (privacy concern) or cap at 50 free files. PikDraw's tool is unlimited, free, runs on every modern browser, and never sees your photos. Perfect for a one-off cleanup of a few hundred shots.
Common use cases
- De-duplicate a phone photo library before backup
- Find recompressed copies of the same stock photo across folders
- Identify near-duplicate product shots in an e-commerce catalogue
- Clean up imported photo libraries after a phone migration
- Audit a design-asset library for accidentally re-saved variants
- Find duplicate icons / textures in a game asset folder
- De-duplicate scanned documents before OCR
How to use this tool
- Upload your folder — Drop any number of JPG / PNG / WebP files. The tool keeps every upload in memory until you clear it — large folders (1000+ photos) work but consume more RAM.
- Set similarity threshold — Lower = stricter. 0–5 catches near-identical files (exact dupes, recompressed copies). 10–15 catches resized / cropped variants. 20+ groups visually similar shots (burst sequences).
- Find Duplicates — The tool computes an average hash (aHash) for every image — a 256-bit fingerprint derived from a 16×16 grayscale grid. Hashes are compared with Hamming distance to group similar images.
- Review the groups — Each group shows the similar photos side by side. Click the × on any image to remove it from the gallery (in-browser only — your disk is untouched).
- Delete duplicates manually — The tool identifies dupes but never deletes files from your disk — browsers can't. Use the visual review to decide which copies to keep, then delete the rest in your file manager.
Who should use this
Anyone with a messy photo library — photographers, designers, marketers, parents. CMS administrators auditing media libraries. Stock contributors checking for accidental re-uploads. Game developers de-duplicating asset folders.
How to get started
Drop your folder, leave threshold at 10, click Find Duplicates. Review the groups, decide which copies to keep, and delete the rest in your file manager.
Best practices
- Threshold 10 is a good starting point — adjust based on results
- Lower the threshold if you see false positives
- Raise it for noisy phone libraries with burst-mode shots
- Always review visually before deleting — perceptual hashes can have false positives
- Process in folders of 2 000–5 000 for very large libraries
- Keep originals — there's no undo on filesystem deletes
- Re-run after every major library change
Pro tips
- Start with threshold 10 — catches genuine duplicates without too many false positives.
- Increase threshold for noisy phone photo libraries with lots of burst-mode shots.
- Decrease for archival deduplication where only exact dupes count.
- The tool is per-session — closing the tab loses the analysis.
- Average hash is fast but not rotation-invariant — rotated copies won't match.
Expert insights
💡 Pro Tip: Start at threshold 10
10 catches most genuine duplicates without false positives. Adjust up for fuzzier matches, down for stricter ones.
💡 Pro Tip: Review before deleting
Perceptual hashes occasionally cluster visually similar but distinct shots. Always eyeball the group before deleting.
💡 Pro Tip: Rotation = no match
If you've rotated copies of the same photo, aHash will treat them as different. Rotate them back to portrait/landscape first.
Limitations to be aware of
- Not rotation- or mirror-invariant
- O(n²) grouping is slow above ~10 000 files
- aHash is faster but less accurate than pHash / dHash
- No automated file deletion (intentional)
- Per-session memory — close the tab and re-upload to re-analyse
- Bounded by browser RAM for very large libraries
Frequently asked questions
- How does the duplicate detection work?
- Each image is downscaled to a 16×16 grayscale grid (256 pixels). The mean brightness is computed; each pixel is encoded as 1 if brighter than the mean, 0 if darker. The result is a 256-bit fingerprint — the average hash (aHash). Two images are considered duplicates when their hashes differ by no more than the threshold bits (Hamming distance).
- Will it catch resized or recompressed copies?
- Yes. aHash is robust to resize and modest recompression because it normalises to brightness on a coarse grid. A 4000 px JPEG and its 800 px WebP downscale typically score a Hamming distance under 5. Heavy recompression or aggressive colour grading can push the distance up — increase the threshold.
- Will it catch rotated or flipped duplicates?
- No. Average hash and perceptual hash both depend on pixel position. A 90° rotated copy of the same photo has a completely different hash. For rotation-invariant matching you need feature-based algorithms (SIFT, ORB) which are too heavy for browser use.
- What threshold should I use?
- Start at 10. Below 5 catches only near-identical dupes (different filename, same bytes after recompression). Above 20 catches visually similar but distinct shots (consecutive burst frames). The optimal threshold depends on your library — review a few groups to calibrate.
- Why doesn't it just compare file hashes?
- MD5 / SHA-1 byte hashes catch only exact-byte duplicates. The moment a file is recompressed, resized, or has metadata changed, the byte hash changes — even though it's visually the same image. Perceptual hashing (aHash, pHash, dHash) compares what the image looks like, not its bytes.
- Can it delete duplicates for me?
- No. Browsers cannot modify the filesystem. The tool identifies and groups duplicates; you delete the unwanted copies in your file manager. This is intentional — automated deletion of photos is dangerous and irreversible.
- Are my files uploaded?
- No. Hashing runs entirely in your browser using the Canvas API. The 256-bit hashes themselves never leave the page. Safe for confidential or unreleased imagery.
- How large a batch can I process?
- Browsers comfortably handle a few thousand images. The hashing step is O(n); the grouping step is O(n²) so 10 000+ files start to feel slow. For massive libraries, process in folders of 2 000–5 000.