What's a good distance threshold?

Under 2 is essentially identical, under 10 is a strong duplicate, 10–15 is similar, and beyond that the images are visually distinct.

Why aHash and not pHash or dHash?

aHash is simple, fast, and good enough to catch most near-duplicates. pHash and dHash are more robust to rotation/cropping but slower.

ahash · perceptual · hamming · local

Detect near-duplicates with perceptual hashing

Computes a 64-bit aHash for each image and reports the Hamming distance between them. Runs locally.

Image Hash Compare

Hash A

—

pick

Hash B

—

pick

Hamming

—

bits flipped

Verdict

—

load both

input · image

Image A

image/* · runs locally · nothing uploaded

Image B

image/* · runs locally · nothing uploaded

output · preview

How aHash works

1. Downscale to 8x8 grayscale.

2. Compute the average luminance.

3. Each cell becomes a bit (1 if >= avg, else 0).

4. 64 bits → 16-char hex hash.

Verdict guide

< 2 bits = identical

< 10 bits = near-duplicate

< 16 bits = similar

≥ 16 bits = different

Quick Start

How it works

Pick an image, tune the options, then download the result. Everything runs in your browser.

01
Load image A
The reference image. It is shrunk to 8x8 to compute the hash.
Files stay in memory as an HTMLImageElement — the original bytes are never uploaded, and EXIF metadata is preserved until you explicitly transform it.
Auploadcanvaslocaldecode
image · bytesA · loaded
```
hash(A) = aHash(8x8)
```
loaded · decodedcanvas · ready
02
Load image B
The candidate. Same 8x8 averaging produces its own 64-bit hash.
Encoder quality is exposed where the format supports it — JPEG and WebP get a 0–1 quality slider; PNG is lossless. Options re-render the canvas immediately.
Boptionslive previewencoderquality
image · tunableB · live
```
hash(B) = aHash(8x8)
```
options applied · liveencoder · live
03
Hamming distance
Count the number of differing bits between the two 64-bit hashes.
Pixel work runs on the Canvas 2D context — drawImage resamples with the browser’s bilinear filter and getImageData/putImageData back per-pixel passes.
bitsrenderCanvas 2Dpreviewpx-perfect
popcount · resampled— · px-perfect
```
popcount(A XOR B)
```
preview · livepx-perfect
04
Verdict
0 means identical. Under 10 is a near-duplicate; over 16 means probably different.
Downloads ship via a hidden anchor with an object URL — no server round-trip, and the file on disk is untouched. The output filename inherits a sensible suffix.
judgedownloadblobclient-sideobject URL
ok · blob— · saved
```
distance → —
```
shipped · ready to shipsuccess

Pro tips

Browser-only image work that scales

Convert with quality control

JPEG/WebP let you pick a 0–1 quality. Default 0.92 is a good compromise of size and visible quality.

PNG isn’t free

Use PNG only for transparency or sharp edges. For photos, WebP is smaller at the same perceived quality.

Strip EXIF for privacy

Photo EXIF often contains GPS, camera model, and timestamps. Strip before sharing public uploads.

Watch the colour profile

Convert to sRGB before web export. Wide-gamut images render oversaturated in non-managed browsers.

AVIF when it’s available

AVIF beats WebP on size at the same quality. Detect support via <picture> with fallbacks.

Lazy-load below the fold

Add loading="lazy" on <img>. The browser handles scheduling — no IntersectionObserver needed.

Use Cases

When to use this tool

Common workflows it speeds up.

De-dupe a library

Find images that are visually the same despite re-encoding. Decoded into a Canvas 2D context — original bytes never leave the tab, so you can process customer-uploaded photos, internal screenshots, or sensitive diagrams without privacy concerns.

px-perfect

CDN cleanup

Spot multiple uploads of the same asset. Encoders expose quality 0–1 where the format supports it (JPEG, WebP); PNG and AVIF stay perceptually lossless. Pick 0.85 for web, 0.95 for archival, 0.7 for thumbnails.

quality-tunable

Crop detection

aHash catches mild crops/resizes that pixel-diff misses. Pair with the EXIF Stripper before sharing user-uploaded photos publicly. Camera serial numbers, GPS coordinates, and timestamps often hide in image metadata and leak more than the picture itself.

format-aware

Stock vs original

Check if a freshly-licensed image already exists in your library. Output preserves colour profile (sRGB) so it renders consistently across browsers, monitors, and print pipelines. Wide-gamut workflows can opt into Display P3 for modern Apple devices.

EXIF-safe

Plagiarism check

Compare two screenshots for visual lift. Re-runs only on option changes, never on window resize, so the canvas stays responsive on large images. drawImage uses the browser's bilinear filter for fast resampling.

client-side

Bot detection

Catch repeated avatars across spammy accounts. Downloads as .png, .jpg, .webp, or .avif with the correct MIME so editors apply native handling on save. The filename inherits a sensible suffix derived from the input.

compression-ready

FAQ

Frequently asked questions

What's aHash?

Average hash: downscale to 8x8 grayscale, threshold against the mean. Cheap and surprisingly robust.

Compared to pHash or dHash?

aHash is the simplest. pHash uses DCT and is more resilient to lighting changes. For tighter accuracy, pre-process the images first.

Will resizing or crop hurt it?

Mild resize is fine. Heavy crop changes the histogram and may push it past 'near-duplicate'.

Does color matter?

Only via luminance. Color-swapped images can still be near-duplicates.

Is anything uploaded?

No. Everything runs locally in your browser.

Suggest a tool

Something missing?

Tell us what you need. We ship the top requests every two weeks.

47 in queue·8 shipped this month

Request queued. Thanks!