ENCRYPTED PICTURE V2 - SUCCESS

Invisible Forensic Watermark Lab
Invisible Forensic Watermark LabStarting engine…

Forensic watermark lab

Give every visitor a visually identical copy of the same photo, each carrying a different invisible ID. When a copy leaks, upload it here to find out whose copy it was.

Workflow

Every step runs inside this page. Nothing is uploaded to a server.

1
Encode — upload a photo, pick visitors and an algorithm, generate one marked copy per visitor.
上传照片 → 为每位访客生成肉眼无差别、但隐藏不同 ID 的副本。
2
Leak — the copy is screenshotted, cropped, compressed or forwarded.
截图、裁剪、WhatsApp / Telegram 转发等。
3
Decode — upload the leaked image, select the photo region, analyse.
上传疑似泄露图片,裁剪照片区域,提取水印并查询访客。
4
Measure — run the Test Lab benchmark to see real detection and false-attribution rates.
所有成功率均为实际运行结果,绝不预设。

Engine

Attribution is not proof of guilt. A match shows which copy the image came from. It does not prove that person leaked it — copies can be passed on. 水印归属 ≠ 实际泄露责任。

Recent analyses

Encode

Generate personalised copies of one photo. Each copy hides a random watermark ID linked to a visitor in your local registry.

1

Upload photo

2

Visitors

Each selected visitor receives a distinct copy with an auto-generated watermark ID. Names never go into the image — only the random ID does.

3

Algorithm

Pick one, or several to compare. Choosing more than one generates every visitor × algorithm combination.

4

Watermark strength

LowHigh 5
较强水印可能提高部分攻击下的识别率,但也可能降低图片质量。Stronger marks survive more abuse but change more pixels; the quality report after generating shows the real cost.
5

Generate

Decode

Forensic image analyser. Upload a suspected leak, select the photo inside it, and extract the hidden watermark.

Suspected image

Detection settings

Detection mode

Algorithm

Test Lab

Automated stress tests on real pixels. Every number here comes from running the encoder and the blind decoder — nothing is simulated or preset.

Benchmark source

Uses the photo loaded in Encode. Tests run on temporary visitors A, B and C with fresh IDs, so your registry is not touched.

Configuration

Algorithms

Strength

LowHigh 5

Scope

Manual real-world tests (T16–T19)

These cannot be simulated. Take a real screenshot or forward the copy through the app, then analyse it in Decode and log the outcome. JPEG compression is never substituted for a platform result.

流程:Encode → Preview for screenshot(或下载后转发)→ 截图 / 保存收到的图片 → Decode → “Log as manual test”。

Threshold calibration

Runs the decoder on random noise to measure how high the detection score gets when no watermark exists. The match threshold must sit well above this noise ceiling.

Registry

The local mapping between watermark IDs, visitors and source photos. Detection looks up IDs here.

Issued copies

One row per watermarked copy. The watermark ID is what is hidden inside the pixels.

Visitors

Source photos

Reports & data

Detection reports, benchmark runs, database backup and the detection key.

Saved reports

Database backup

Stored in this browser's IndexedDB. Browsers can clear it, and it does not sync between devices — export regularly.

Detection key

A 256-bit secret seeds the watermark layout and signs every payload (HMAC). Images marked with one key cannot be decoded with another.

Fingerprint
Created
Test-grade key storage. The key lives in local browser storage next to the registry. That is fine for evaluating algorithms, but a production deployment must keep keys and the ID mapping on a trusted server. 本地密钥不适合直接作为生产环境的长期安全方案。