Whitepaper: Long-Term Proof of Authorship for Digital Images (Perceptual Fingerprints + Human-Readable Records)

Today I published a whitepaper on my website that proposes a practical framework for long-term proof of authorship for digital images—without needing to store image files on-chain or rely on fragile hosting links.

If you’ve been around NFTs for a while, you’ve probably seen the problem: the token may survive, but the referenced image can disappear when a marketplace shuts down, a gateway changes, or hosting quietly breaks. And even outside NFTs, the day-to-day reality is that images are shared as web exports: resized, recompressed, watermarked, and often stripped of EXIF/XMP metadata. A byte-exact hash is great for integrity checks, but it fails as soon as the image is exported differently.

My proposal is simple: register a compact, non-visual proof record on-chain that remains useful across typical derivatives.

The core idea (in plain terms)

Instead of storing the image on-chain, store:

• A perceptual fingerprint bundle (designed to match common derivatives such as resizing, recompression, light watermarks, and metadata stripping)
• A structured, human-readable description of what’s visible (not marketing copy)
• Authorship metadata (creator identifier + timestamp)
• Optional rights events (ownership transfers, licenses, editions)

This creates a durable “anchor” that can still be used when the file is encountered in a different resolution or stripped-down web version.

What this is (and isn’t)

This framework is designed for typical distribution variants. Matching is threshold-based and probabilistic, so it is not meant to guarantee recognition under heavy editing, compositing, style transfer, or adversarial transformations intended to evade detection.

In other words: it won’t prevent copying. It aims to make attribution and verification more durable.

Why “discovery” matters

A registry only becomes truly useful when other people can check it.

That’s why the whitepaper also includes a pragmatic discovery model:

• On-chain stores the record as a source of truth.
• Off-chain indexers enable similarity search (nearest-neighbor retrieval), because that is not practical to do on-chain at scale.
• Indexers aren’t trusted parties: they return candidates; the results are then verified directly against the chain.

This makes it possible for buyers, platforms, and licensees to check claims—useful in cases like stolen images being listed on marketplaces.

Optional “Tier 2” confirmation (when you need it)

For rare overlap cases or disputes, the framework allows an optional external reference to a canonical file (hosted elsewhere), accompanied by a cryptographic hash. This is non-normative: the on-chain record remains valid without it, but it enables file-level confirmation when needed. The tradeoff is clear: availability becomes the responsibility of the rights holder.

Read the whitepaper

• Web version: https://www.arjenroos.com/whitepaper-long-term-image-authorship/
• PDF download (v1.1): (use the download link at the top of the page)

If you build in Web3, rights tech, or creator tooling and want to discuss implementation paths (PoC, schema design, indexer approach, or attestation/KYC integration), feel free to reach out.