Bakau in one page

Image fingerprinting and provenance infrastructure: finding every copy of an image, recording who made what, and making rights legible to software and buyers. Early, and run by its founder, Joseph Sherrier (Georgia Tech).

Print this page or save it as PDF; the navigation drops away.

Built today

• Copy matching that survives resizing, re-encoding, crops, rotations, mirrors and borders: perceptual hashes with private region checks, plus a copy-detection descriptor confirmed by geometry. Python, on the CPU, offline.
• A catalogue audit that runs where the images are: groups copies, reads creator and rights metadata, flags gaps and conflicts, and writes HTML, CSV and JSON reports.
• A timestamped record: a Merkle root over every record, filed with OpenTimestamps.
• Fingerprint and discard: no image is ever stored.

Benchmark

On 200 public-domain photographs with 19 edits each, the audit matcher found 99% of copies and 97% of crops, with 0 false matches in 3,200,000 unrelated pairs (2026-10-10). Caveat: a small, synthetic test on stock-style photographs; real catalogues are harder. It looks for copies, not look-alikes, can occasionally join two photographs of one scene by one photographer, and never says who made an image. Full method: https://bakau.org/benchmarks

Working with Bakau

• Software partners: license the matching engine (PolyForm Noncommercial 1.0.0 for noncommercial use; commercial use needs a licence), and have it benchmarked on your own test set.
• Archives, agencies and marketplaces: a free diagnostic on about 1,000 images, run on your systems, then a full audit if it is useful.
• Creators and associations: creators never pay. Collective licensing is an aim, not a feature.

Limits

Evidence, never rulings on ownership, and not legal advice. Reverse search can show where and when an image appeared, never who made it. Metadata is quoted as found, unverified, and can be wrong.

Contact

outreach@bakau.org · https://bakau.org