Glossary · Near-duplicate photo
What is a near-duplicate photo? Similar shots that file checks miss
Short answer
A near-duplicate photo is a picture that looks almost the same as another one but is stored as a different file. Two shots of the same pose a second apart, or a WhatsApp copy shrunk from your original, are near-duplicates. File-hash tools miss them, so finding them needs visual comparison, such as a perceptual hash or an AI model.
Last updated
What counts as a near-duplicate?
| Example | Why it differs | Exact duplicate? |
|---|---|---|
| Three taps of the shutter on one pose | Tiny movement, different noise | No |
| Your photo re-sent through a chat app | Re-compressed, smaller size, metadata removed | No |
| A photo and its edited version | Crop, filter or brightness change | No |
| Screenshot of a photo | New pixels, status bar added | No |
Each row produces a new file with a new hash, so a byte comparison sees two unrelated files. To a person they look like the same picture.
Screenshots of photos are a common blind spot. They carry a status bar and sometimes a chat bubble, yet the image in the middle is the same as one already in your camera roll.
How is a near-duplicate detected?
Tools reduce each photo to a compact visual fingerprint. A perceptual hash shrinks the image to a tiny greyscale grid and records its pattern of light and dark; similar images give hashes that differ in only a few bits. AI cleaners use image embeddings instead, which also cope with small crops and colour changes.
Capture time is a second signal. Shots taken seconds apart in the same place are likely one moment, while the same view a week later is a separate memory and should stay.
Thresholds matter. A tight threshold groups only shots that are almost pixel-identical; a loose one starts lumping together different photos of the same room. Good tools combine visual similarity with time and let you split or merge a group by hand.
Time stamps also protect deliberate repeats. A similar view taken months apart is treated as a separate photo, so seasonal or yearly shots of the same spot are not grouped together.
Should every near-duplicate be deleted?
No. Near-duplicates are a decision, not an automatic delete. Within a group, one shot is usually sharper, better exposed or better framed, and that one deserves to stay. An edited version may matter more than the original. A sensible rule is to keep one per moment and review the rest side by side. Our guide on deleting similar photos on Android covers the free options.
GalleryJury is built around this case. The coming-soon Android app groups shots of one moment, marks the best one to keep and lists the rest for your OK. Favourites and photos from the last 48 hours are left alone.
Re-sent copies are the clearest near-duplicates to remove. When your own camera original is on the phone, the smaller copy that came back through a group chat adds nothing.
Frequently asked questions
Why doesn't Files by Google find my similar photos?
Its Duplicate files card matches identical files. Similar shots are different files, so they need a visual tool, though its blurry-photo card can catch some weak shots.
Are lower-resolution copies near-duplicates?
Yes. A WhatsApp or email copy is re-encoded at a smaller size, so the bytes change even though the picture looks the same.
Do near-duplicates take much space?
Burst-style groups add up fast. A 12 MP JPEG is often 3 to 5 MB, so ten extra shots of one pose can cost 30 to 50 MB.
Sources
- Wikipedia: Perceptual hashing (checked 31 October 2026)
- Files by Google Help: Delete duplicate files (checked 31 October 2026)
Related
More in Photo cleanup terms
GalleryJury is coming soon to Android
It keeps the best shot from every burst and clears the clutter around it. 50 free photos to try, then $5 for 5,000 photos or $15 for 20,000. Pay once, nothing renews. No subscription.