DCT-Watermark-Camera

Forensic watermark camera — 3 visually-lossless algorithms (QIM-DCT/DWT/spread spectrum), zero color loss, RS error correction, survives JPEG re-sharing

取证水印相机：QIM-DCT/DWT/扩频三种视觉无损水印算法，色彩零损失，Reed-Solomon 纠错，JPEG 转发后仍可提取

---

Table of Contents / 目录

• English
• 中文

---

English

Algorithms

Three visually-lossless watermark algorithms, switchable in the UI. All embed into the luminance channel only (via a reversible YCoCg transform), leaving chroma untouched:

Algorithm	Technique	Properties
QIM-DCT (color-preserving, default)	8×8-block DCT mid-frequency QIM quantization	Zero chroma loss, PSNR ≈ 55 dB, large capacity
DWT (Haar)	Haar wavelet HL-subband QIM	Strongest noise tolerance (noise ≤ 10), fastest
Spread spectrum	PN-sequence spreading + centered correlation detection	Most robust (noise ≤ 15, survives JPEG Q70)

Shared by all three:

• Reed-Solomon error correction (fail-closed) — decoding failure returns an empty string: "no watermark" is a reliable verdict, never garbage
• Header protection — the length prefix is stored in 3 copies with per-bit majority voting: no single-bit point of failure
• Measured JPEG robustness — the default algorithm survives re-compression down to Q60 (typical chat-app compression), so the watermark remains extractable after the photo is re-shared

Features

• Forensic capture — the shutter automatically collects timestamp + GPS coordinates + device ID (ANDROID_ID / identifierForVendor) and embeds them as an invisible watermark; the photo is saved to the system gallery as lossless PNG
• Verification — pick any image from the gallery; structured display of algorithm / capture time / coordinates / device; every registered algorithm is tried automatically, no manual matching
• iPhone-style camera layout — full-screen viewfinder, bottom control bar (gallery thumbnail / shutter / settings)
• Non-blocking UI — all pixel-level work (conversion, embedding, extraction, PNG encoding) runs in background isolates

How It Works & Honest Limitations

• The watermark lives in frequency-domain coefficients (8×8 DCT blocks / Haar subbands). It survives the app's own lossless PNG round-trip — which is exactly why the app saves PNG. Measured JPEG survival is down to Q60; stronger compression or repeated re-encoding can still damage it.
• Device identifier is Settings.Secure.ANDROID_ID on Android (stable until factory reset; per-app on Android 8+) and identifierForVendor on iOS (changes on reinstall). It identifies a device, not a person.
• Coordinates require location permission; if denied, the watermark simply omits the g segment.
• Not interoperable with the original Android app: the original Java implementation is preserved in git history (commits up to e0073cf) but is no longer selectable in the UI.
• iOS builds require macOS + Xcode; this repository is developed on Windows, so only the Android target is built and tested here.

Build & Run

git clone https://github.com/Mal-Suen/DCT-Watermark-Camera.git
cd DCT-Watermark-Camera
flutter pub get
flutter run                    # or: flutter build apk --release
flutter test                   # 11 tests: round-trips, clean-image verdict, JPEG robustness, UI smoke

Requires Flutter 3.29+; Android builds need JDK 17, NDK 27.0.12077973, Kotlin 2.2.0.

Usage

1. Launch the app — the full-screen camera viewfinder starts. Tap the settings button (bottom-right of the control bar) to pick an algorithm if you don't want the default.
2. Press the shutter — the app collects time/position/device, embeds the watermark, and saves the photo to the gallery (a toast shows the embedded string); the bottom-left thumbnail updates to the latest capture.
3. Tap the bottom-left thumbnail to open the verification sheet, pick that photo from the gallery, tap 提取水印 — the structured fields (algorithm, capture time, coordinates, device) are displayed.

Project Structure

DCT-Watermark-Camera/
├── lib/
│   ├── main.dart               # App entry: iPhone-style camera UI, capture pipeline, verify sheet
│   ├── evidence_collector.dart # Evidence model: assemble & parse the watermark string
│   ├── image_codec.dart        # ui.Image <-> bitmap bridging, PNG codec, thumbnails
│   └── algorithm/              # Watermark algorithms (pure Dart, no UI deps)
│       ├── algorithm.dart      # Barrel export + defaultAlgorithms() registry
│       ├── dct_qim_color.dart  # QIM-DCT color-preserving (default)
│       ├── dwt_algorithm.dart  # DWT domain
│       ├── spread_spectrum.dart# Spread spectrum
│       ├── color_space.dart    # Reversible RGB <-> YCoCg transform
│       ├── reedsolomon.dart    # RS error correction
│       └── ...                 # Legacy compat algorithm, math utilities
├── test/                       # Unit & end-to-end tests (incl. JPEG robustness regression)
├── bin/                        # Diagnostic scripts (dart run bin/...)
└── android/ ios/ web/ windows/ # Platform projects

---

中文

算法

三种视觉无损水印算法，UI 可切换。全部只嵌入亮度通道（经 YCoCg 可逆变换），色度完全不动：

算法	技术	特性
QIM-DCT 保色（默认）	8×8 块 DCT 中频 QIM 量化	色彩零损失，PSNR ≈ 55 dB，容量大
DWT 域	Haar 小波 HL 子带 QIM	抗噪最强（noise ≤ 10），速度最快
扩频	PN 序列扩频 + 中心化相关检测	鲁棒性最强（noise ≤ 15，JPEG Q70 存活）

三算法共享：

• Reed-Solomon 纠错（fail-closed）——解码失败返回空串："无水印"是可靠判定，绝不返回乱码
• 头部保护——长度前缀 3 份副本按位多数表决：无单 bit 单点故障
• JPEG 鲁棒性实测——默认算法存活至 Q60（聊天工具标准压缩档），照片经转发后水印仍可提取

特性

• 取证拍照——快门自动采集 拍摄时间 + GPS 经纬度 + 设备唯一标识（ANDROID_ID / identifierForVendor），嵌入不可见水印，以无损 PNG 保存到系统相册
• 验水印——相册选图提取，结构化展示 算法/拍摄时间/坐标/设备；自动遍历全部注册算法，无需手动匹配
• iPhone 相机布局——全屏取景 + 底部控制条（相册缩略图 / 快门 / 设置）
• UI 不阻塞——全部像素级工作（转换、嵌入、提取、PNG 编码）在后台 isolate 执行

工作原理与诚实声明

• 水印承载在频域系数（8×8 DCT 块 / Haar 子带）。它能完好通过本 App 自身的无损 PNG 往返——这正是 App 存 PNG 的原因。JPEG 存活实测至 Q60；更强的压缩或反复重编码仍可能损坏水印。
• 设备标识：Android 用 Settings.Secure.ANDROID_ID（恢复出厂前恒定；Android 8+ 按应用隔离），iOS 用 identifierForVendor（卸载重装会变）。它标识的是设备，不是人。
• 经纬度需要定位权限；被拒绝时水印自动省略 g 段。
• 与原 Android 版不互通：原 Java 实现保留在 git 历史（e0073cf 及之前），但已不在 UI 可选列表中。
• iOS 构建需要 macOS + Xcode；本仓库在 Windows 上开发，仅构建和测试 Android 目标。

构建与运行

git clone https://github.com/Mal-Suen/DCT-Watermark-Camera.git
cd DCT-Watermark-Camera
flutter pub get
flutter run                    # 或：flutter build apk --release
flutter test                   # 11 项测试：往返、干净图判定、JPEG 鲁棒性、UI 冒烟

要求 Flutter 3.29+；Android 构建需 JDK 17、NDK 27.0.12077973、Kotlin 2.2.0。

使用流程

1. 启动 App——全屏相机取景。不用默认算法的话，点底部控制条右侧的设置键选择算法。
2. 按快门——App 采集时间/定位/设备ID，嵌入水印，保存到相册（Toast 显示嵌入内容）；底部左侧缩略图更新为最新拍摄。
3. 点底部左侧缩略图打开验水印面板，从相册选那张照片，点提取水印——结构化字段（算法、拍摄时间、坐标、设备）即被展示。

项目结构

DCT-Watermark-Camera/
├── lib/
│   ├── main.dart               # 应用入口：iPhone 相机布局、拍摄管线、验水印面板
│   ├── evidence_collector.dart # 取证证据模型：水印串组装与解析
│   ├── image_codec.dart        # ui.Image 与位图互转、PNG 编解码、缩略图
│   └── algorithm/              # 水印算法库（纯 Dart，无 UI 依赖）
│       ├── algorithm.dart      # 导出入口 + defaultAlgorithms() 注册表
│       ├── dct_qim_color.dart  # QIM-DCT 保色（默认）
│       ├── dwt_algorithm.dart  # DWT 域
│       ├── spread_spectrum.dart# 扩频
│       ├── color_space.dart    # RGB 与 YCoCg 可逆变换
│       ├── reedsolomon.dart    # RS 纠错
│       └── ...                 # 旧版兼容算法、数学工具
├── test/                       # 单元与端到端测试（含 JPEG 鲁棒性回归）
├── bin/                        # 诊断脚本（dart run bin/...）
└── android/ ios/ web/ windows/ # 各平台工程

---