Loading page
About this tool
Find exact and near-duplicate images using perceptual hashing. Supports batch upload, similarity scoring, and group management.
Duplicate Image Finder compares a batch of images with two 64-bit perceptual hashes — a difference hash and an average hash computed on greyscale — and groups any pair whose averaged similarity reaches 80% or more. Because it matches on visual structure rather than file bytes, it catches a resized, recompressed or renamed copy that a checksum comparison would treat as a different file. Photographers and anyone cleaning a photo folder get similarity-scored groups, a keep-best action, and a CSV report.
Open Duplicate Image Finder on AltFTool — it loads instantly in your browser.
Drop photos on the 'Drop images here or click to browse' zone — multiple images at once, including HEIC/HEIF; non-image files are skipped with a notice.
Press 'Find Duplicates' to compare every pair on 64-bit difference and average hashes, grouping matches at 80% or higher averaged similarity.
Review each group's '% match' badges (95%+ Exact, 85-94% Near), press 'Keep Best' to drop the lower-resolution copies, or 'Export CSV' for duplicate-report.csv.
Each pair is scored on both a difference hash and an average hash and the two are averaged, so a match does not hinge on a single algorithm's blind spot.
Pairs at 95% and above are labelled exact, 85 to 94% are labelled near, so you know which groups need a look before you delete.
Keep-best removes the flagged copies from the working list only; your files on disk are untouched, and you can remove individual entries instead.
Each image is shrunk to an 8×8 greyscale grid and reduced to a 64-bit fingerprint — the difference hash records whether each pixel is brighter than its right-hand neighbour, the average hash whether each pixel is brighter than the image mean. Two fingerprints are compared by Hamming distance, so similarity is the percentage of the 64 bits that agree.
Grouping starts at 80% averaged similarity. In practice 95% and above is effectively the same picture, 85 to 94% is a near-duplicate such as a re-crop or a different export, and anything in the low 80s is worth eyeballing before you act on it.
Yes — that is the point of a perceptual hash. Scaling, JPEG recompression and small colour shifts barely change the 64-bit fingerprint, whereas a file checksum changes completely, which is why a hash-based deduplicator misses these entirely.
No. Images are decoded and hashed in your browser using a canvas, and only the resulting fingerprints are compared, so nothing leaves the machine even for a large batch.