aHash, dHash and pHash: How the Three Perceptual Fingerprints Differ
By the Ingenious AI team · Published: November 14, 2026 · Reviewed: November 14, 2026 · 8 min read
The short answer
ProvenanceGuard computes all three: aHash from pixel averages, dHash from horizontal gradients, and pHash from the low-frequency DCT structure. Each notices different changes, so the registry combines them into a 192-bit comparison for recovery, with a match at a combined distance of 36 bits or fewer. Using three weak-ish witnesses that fail differently beats relying on one.
Asking one fingerprint to catch every kind of edit is like asking one lock to defeat every kind of key; the design answer is several, disagreeing honestly.
The honesty of the method is in its stated limits. This is fingerprint-only soft binding: ProvenanceGuard embeds no invisible watermark, so recovery can only find images that were registered before they travelled, and only while their pixels stay within threshold of the original. Heavy cropping, collage, or aggressive filtering can push a copy beyond matching, and the lookup then says so rather than forcing a nearest guess. A tool that always found something would be manufacturing provenance, which is the opposite of the point.
What is the three perceptual hashes? aHash, which thresholds pixels against the image average; dHash, which encodes horizontal brightness gradients; and pHash, which encodes low-frequency DCT structure. Together they describe an image's appearance from three independent angles.
AHash, dHash and pHash: the plain-English version
Recovery exists for the copy that comes back to you with nothing attached: no metadata, a new filename, dimensions changed by two platform hops. The Recovery Registry lookup fingerprints the uploaded copy perceptually, using aHash, dHash and pHash together, and searches the registry for a record within the stated threshold: a combined distance of at most 36 differing bits out of 192 compared, or 12 out of 64 for older records that carry only the legacy average hash. A match returns the original exhibit, its creator and disclosure declarations, and its registration time.
The registry model also disciplines what a match means. A fingerprint match says this image is the registered image or a close descendant of it; it does not say the person showing it to you owns it, and it does not say the disclosure attached to it was correct. Those judgements stay human. What the registry removes is the argument about identity and timing: which file, registered when, carrying which declaration. Most real disputes shrink considerably once those three facts stop being contestable.
Reading the result properly
The registry is the part that survives distribution. At sealing, the asset's SHA-256, its sealed copy's SHA-256, and its perceptual fingerprints are stored against a public exhibit ID. Anyone with that ID, or with the file's hash, can open the public verification page or query the verification API and confirm that this exact file, or a near copy of it, was registered, when, and with what declared creator and disclosure. Verification pages are public by design: a record only a vendor can check would not settle disputes.
Two kinds of fingerprint do two kinds of work. The SHA-256 is exact: change one byte and it changes completely, which makes it perfect for proving a copy is bit-for-bit the registered file and useless for recognising a resized repost. The perceptual fingerprints, computed from the pixels rather than the bytes, are approximate on purpose: they stay close when a copy is resized, re-encoded or lightly edited, and drift as edits accumulate. A good registry stores both and is honest about which kind of match it is reporting.
Mistakes that quietly break the record
Pipelines break provenance in boring places: a design tool exporting without metadata, a CDN recompressing on the fly, a social scheduler re-encoding before upload, an archive converting everything to one format. Each is a transform, and each transform is testable. The lineage tool's local transforms and the vault battery's transforms exist precisely so a team can run its own pipeline's steps against a sealed sample and watch which step removes what, instead of discovering it from a published asset that no longer verifies.
Format details decide more provenance outcomes than policy documents do. JPEG carries XMP in application segments that re-encoding discards; PNG carries text chunks that many converters drop silently; EXIF, written by cameras, is routinely stripped by platforms and says little about AI generation in any case. ProvenanceGuard works with JPEG and PNG because their structures are where its inspection, XMP embedding and fingerprinting are defined and tested, and it inspects all three metadata territories, XMP, EXIF and PNG chunks, rather than assuming any one of them is where the truth lives.
When free is enough and when paid depth earns its place
ProvenanceGuard is built around those layers, and its wording is deliberately narrow. The free job inspects one JPEG or PNG, embeds a signed ProvenanceGuard assertion in XMP, produces a visibly labelled copy, registers the SHA-256 and perceptual fingerprints, and gives the asset a public verification page. The assertion is HMAC-signed and labelled accurately everywhere as ProvenanceGuard's own assertion; it is not a C2PA manifest and is never described as one.
Provenance is the record of where a piece of content came from and what has happened to it since: who or what created it, when it was registered, and which copies descend from the original. For an AI-generated image that record matters because the file itself is easy to copy, rename, crop and re-upload, and none of those actions leaves a reliable trace inside a filename. A provenance record kept outside the file, in a registry keyed to the file's fingerprints, survives exactly the handling that erases the signals embedded in the file.
Five actions to take
- Read the three distances separately before reading the combined one
- Treat a single-hash record as legacy data and re-seal important assets to gain all three
- Re-check one published asset through the Recovery lookup so you know the stripped-copy path works before you need it
- Seal the final export of the asset, with the creator and disclosure declared in your own accurate words
- Download and keep both copies the seal produces: the sealed copy as the record, the labelled copy for publication
A worked example: A support team testing why a cropped copy failed to match
The crop barely moved the pHash but pushed the dHash and aHash far; reading all three distances showed the copy was a genuine heavy edit rather than a different image, and the team recorded it as a lineage step instead of a failed lookup.
File the pattern, not just the outcome: a dated record, a measured match, and wording that never claimed more than the record holds. Repeated across a catalogue, that pattern is what turns provenance from a feature into a reputation.
FAQ
Why not just use the strongest single hash?
Because each hash is blind to something: averages miss rearrangement, gradients miss flat changes, frequency hashes miss local edits. Combined distance over 192 bits compared is the product's stated recovery method precisely because the three fail differently.
Can anyone verify a record, or only the owner?
Anyone. Verification pages are public by design, the single-hash lookup is a public API, and an exhibit ID or a file hash is enough. A record only its owner could check would not settle a dispute with a stranger.
Is ProvenanceGuard's assertion a C2PA manifest?
No. It is ProvenanceGuard's own signed assertion, embedded as XMP and HMAC-signed, and labelled that way everywhere. The inspection detects genuine C2PA markers when a file carries them, but the two records are never conflated.
What formats and sizes are supported?
JPEG and PNG, up to 25 MB per file. Those are the formats the inspection, XMP embedding and fingerprinting are built and tested against; other formats are refused rather than approximated.