Vérifier
Demander un certificat
XENLOOK C2PAVérifier

Pilote · revue C2PA · Corée mars 2027 · Japon juin 2027 · droit coréen

Copyright C-2026-036487 · Demande de brevet 10-2026-0100026 (pas encore enregistrée)

Mentions légales et politiques →
Course contents
Chapter 6 of 10

Chapter 6The stamp you see and the stamp you don't

Photos posted to social media are almost always shrunk and saved again. In this chapter you learn how a visible watermark differs from an invisible one, then test for yourself what survives when a photo is changed.

  • Official session name · Invisible watermark design analysis and robustness tests on field data
  • 6 practice steps
  • Some steps need you to sign in
What you will learn in this chapter
  • You know the strengths and weaknesses of visible and invisible watermarks.
  • You test what happens to each layer when a photo is shrunk or saved again.
  • You can put your test results in a table without overstating them.

The idea in plain words

Like the picture on a banknote and its hidden watermark
Anyone can see the picture on the front of a banknote, but it is also easy to imitate. The hidden image you only see against the light stays out of sight, yet it stops forgers. A visible watermark is the front picture; an invisible watermark is the hidden image. Neither is perfect, so they are used together.

A visible watermark is a logo or text placed on top of the photo. Anyone notices it at once, but it is easy to crop out or erase.

An invisible watermark spreads changes too small for the eye to notice across the whole photo. It is hard to remove, but heavy shrinking, rotating or recompressing can weaken it.

So no single layer is relied on. If the C2PA record is stripped, the hidden mark may remain; if the hidden mark is weakened, a similar fingerprint can still find the source. This is why the five layers in chapter 3 exist.

Know this site's measurement status honestly. The hidden mark is weak against a 5° rotation. And as of 17 October 2026 the transformation tests (T1–T7) have not been measured again, so the watermark results in the public table read “pending verification.”

The seven transformation tests are: T1 original, T2 manifest stripped, T3 saved again as JPEG at quality 70, T4 shrunk, T5 cropped, T6 rotated 5°, T7 re-encoded. The tests in this course use the same names.

See the picture

  1. Originalall five layers present
  2. Record strippedmanifest strip
  3. Saved againJPEG quality 70
  4. Found againchecked with the layers left
This is what the three-stage fallback check does. Even in a photo whose record was stripped and that was saved again, it finds the source with the layers that remain.

Try it yourself

Work from the top, one step at a time. After each step, check that your screen matches “What you should see.”
  1. Read the hidden mark page and look at the “invisible stamp experiment.” Compare the left and middle photos, then look at the “magnifier” photo on the right.
    What you should see The left and middle photos look the same to the eye; only the right one reveals the embedded pattern.
    Open the hidden mark pageNo sign-inPartly in Korean
  2. On the practice page, in the “three-stage fallback check,” press “strip + JPEG q70 → run the three-stage check.” See what remains in a photo whose record was stripped and that was saved again at quality 70.
    What you should see Stages 1 and 2 show “abstain,” and the final verdict reads “restored from the registry (PDQ).”
    Open the practice pageNo sign-inPartly in Korean
  3. Make three variants of your chapter 2 photo: ① shrink it to half size, ② save it again in a photo app, ③ crop a little off the edge. Put each one into the practice page and write down its hash.
    What you should see All three variants have a hash different from the original.
    Open the practice pageNo sign-inPartly in Korean
  4. Apply the same changes to a photo with a hidden mark that your mentor prepared, then put it into “Server reader — six lines, measured one by one” (members) lower on the verify page to see whether the hidden mark survived.
    What you should see The result shows how close the match is, for example “mark match ○○% · threshold: 65% or more.”
    Open the verify pageSign-in neededWith your mentorPartly in Korean
  5. Read the T1–T7 table on the robustness test page. Copy any cell marked “pending verification” exactly as “pending verification.”
    What you should see You see the seven rows T1–T7 and their measurement status.
    Open the robustness testsNo sign-inPartly in Korean
  6. Make a summary table of your test results. The rows are the changes (original, shrunk, saved again, cropped); the columns are hash, hidden mark, fingerprint and C2PA record. In each cell write one of: kept / changed / gone / not measured.
    What you should see What you saw yourself and what you read on a page are kept apart.
    No sign-in
The part you do with your mentor · The watermark mentor prepares a practice photo that carries a hidden mark and shows you how to read the match percentage in the result. You also learn not to call something “genuine” from a single number.

Check your work

If you can answer “yes” to everything below, you have finished this chapter.
  • Can you name one strength and one weakness each of visible and invisible watermarks?
  • Did you write down the final verdict of the three-stage fallback check?
  • In your summary table, did you keep “what I saw” apart from “what I read”?

Easy to mix up

  • Common misunderstandingAn invisible watermark survives any change.
    What is actually trueHeavy shrinking, rotating or recompressing can weaken it. This site's hidden mark is also weak against a 5° rotation.
  • Common misunderstanding“Abstain” means the file is fake.
    What is actually trueIt means that stage could not confirm anything. The check holds back on purpose to avoid false alarms.
  • Common misunderstandingA cell that was not measured can be written down as “pass.”
    What is actually trueAnything not measured is written as “not measured” or “pending verification.” That is an honest record.

Review questions

  1. Question 1What is the biggest weakness of a visible watermark?Show answerHide answer
    It is easy to crop out or erase.
  2. Question 2In the three-stage fallback check, stages 1 and 2 show “abstain,” but stage 3 restored the source. What does that mean?Show answerHide answer
    Neither the manifest nor the hidden mark could confirm it, but the registered similar fingerprint (PDQ) found the source again.
  3. Question 3The watermark result in the public table reads “pending verification.” How should you write it in your report?Show answerHide answer
    Write “pending verification” as it is. Do not turn it into a pass or a fail.

Key terms

Visible watermark
A logo or text placed on top of a photo.
Robustness
How well information survives when a file is changed.
Re-encoding
Saving a file again in a new format or with new settings.
Abstain
Stopping instead of forcing a verdict when something cannot be confirmed.
Fallback
Moving on to the next method when the one before it does not work.

Practice log

Print it and write by hand, or copy the same items into a document. Once your mentor has checked it, you are done.
Chapter 6 practice logThe stamp you see and the stamp you don't
Practice date
Written by
Mentor check
  1. What you saw in the “invisible stamp experiment”
  2. The three-stage fallback check result (stages 1, 2 and 3, and the final verdict)
  3. The hashes of your three variants (same as the original / different)
  4. The hidden mark match you checked with your mentor
  5. Your summary table of test results
  6. The cells left as “pending verification” or “not measured,” and why