🛡️ Digital Asset Protection System

Upload → Watermark → Store → Detect → Compare → Classify → Display

An end-to-end MVP for protecting digital image assets using invisible steganographic watermarking, AI-powered feature extraction (MobileNetV2), and perceptual hash similarity detection — deployed on Google Cloud Run.

---

Architecture Overview

┌─────────────┐     HTTP/multipart      ┌──────────────────────────────────┐
│  Streamlit  │ ─────────────────────→  │        FastAPI Backend           │
│  Frontend   │ ←─────────────────────  │                                  │
│  :8501      │       JSON response     │  /upload-original                │
└─────────────┘                         │  /check-image                    │
                                        │  /assets                         │
                                        └───────────┬──────────────────────┘
                                                    │
                        ┌───────────────────────────┼────────────────────┐
                        ▼                           ▼                    ▼
               ┌─────────────────┐      ┌─────────────────┐    ┌────────────────┐
               │  Stegano (LSB)  │      │  MobileNetV2    │    │   SQLite DB    │
               │  Watermarking   │      │  Embeddings     │    │  assets.db     │
               └─────────────────┘      └─────────────────┘    └────────────────┘

Decision Logic

Watermark Valid  ─────────────────────────────► Authentic Content ✅
Watermark Missing + High Similarity (>75%)  ──► Unauthorized Copy ⚠️
Watermark Corrupted  ────────────────────────► Tampered Content ⚠️
Low Similarity (<40%)  ──────────────────────► Different Content ❌

---

File Structure

digital-asset-protection/
├── backend.py          # FastAPI app (all core logic)
├── frontend.py         # Streamlit UI
├── requirements.txt    # Python dependencies
├── Dockerfile          # Container definition
├── start.sh            # Process launcher script
└── README.md           # This file

---

Local Setup & Run

Prerequisites

• Python 3.10 or 3.11
• pip

1. Install dependencies

# Clone / download the project folder, then:
cd digital-asset-protection

# (Optional) create a virtual environment
python -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate

# Install CPU-only PyTorch first (smaller, faster install)
pip install torch==2.3.0+cpu torchvision==0.18.0+cpu \
    --index-url https://download.pytorch.org/whl/cpu

# Install remaining dependencies
pip install -r requirements.txt

2. Start the backend (Terminal 1)

python backend.py
# → FastAPI running at http://localhost:8000
# → Auto-downloads MobileNetV2 weights on first run (~14 MB)

3. Start the frontend (Terminal 2)

streamlit run frontend.py
# → Opens http://localhost:8501 in your browser

4. Test via API directly (optional)

# Register an original image
curl -X POST http://localhost:8000/upload-original \
  -F "file=@your_image.jpg"

# Check a suspected image
curl -X POST http://localhost:8000/check-image \
  -F "file=@suspected_image.jpg"

# List all registered assets
curl http://localhost:8000/assets

---

Example Output

{
  "status": "success",
  "watermark_detected": false,
  "watermark_status": "missing",
  "matched_content_id": "3f8a2b1c-...",
  "matched_filename": "logo.png",
  "hash_similarity": 85.9,
  "ai_similarity": 92.3,
  "decision": "Unauthorized Copy ⚠️",
  "confidence": 89.8,
  "explanation": "ImageHash Similarity: 85.9% | AI Similarity: 92.3% | Final Decision: Unauthorized Copy ⚠️ | Confidence: 89.8%"
}

---

Deploy to Google Cloud Run

Prerequisites

• Google Cloud SDK installed
• Docker installed
• A GCP project with billing enabled

Step 1 — Authenticate & set project

gcloud auth login
gcloud config set project YOUR_PROJECT_ID

Step 2 — Enable required APIs

gcloud services enable run.googleapis.com containerregistry.googleapis.com

Step 3 — Build & push Docker image

# Build
docker build -t gcr.io/YOUR_PROJECT_ID/dap-system .

# Push
docker push gcr.io/YOUR_PROJECT_ID/dap-system

Step 4 — Deploy to Cloud Run

gcloud run deploy dap-system \
  --image gcr.io/YOUR_PROJECT_ID/dap-system \
  --platform managed \
  --region us-central1 \
  --allow-unauthenticated \
  --memory 2Gi \
  --cpu 2 \
  --port 8501 \
  --set-env-vars="STREAMLIT_SERVER_HEADLESS=true"

Step 5 — Update API_BASE in frontend.py

After deployment, grab the Cloud Run service URL and update frontend.py:

# frontend.py line ~20
API_BASE = "https://dap-system-xxxx-uc.a.run.app"  # ← your Cloud Run URL

Then rebuild and redeploy.

Alternative: Deploy to Railway / Render

Both platforms support Docker deployments via GitHub. Push this folder to a repo, connect to Railway or Render, set PORT=8501, and it deploys automatically.

---

Scalability Roadmap

This MVP uses SQLite and local disk. Here's how to scale to an internet-scale production system:

1. Cloud Object Storage

• Replace local file I/O with Google Cloud Storage (GCS) or AWS S3
• Store watermarked images as objects; use signed URLs for secure, time-limited access
• Add a CDN (Cloud CDN / CloudFront) for global low-latency delivery

2. Vector Database for Embeddings

• Replace numpy cosine search with FAISS (open-source, in-process, billions of vectors)
• Or Pinecone / Weaviate / Qdrant for managed ANN search at scale
• Sub-millisecond similarity search over millions of registered assets

3. Real-time Ingestion Pipelines

• Cloud Pub/Sub + Cloud Run Jobs for async, event-driven watermark embedding
• Vertex AI Endpoints for scalable, GPU-backed MobileNet inference
• Redis for embedding cache and rate limiting per user/API key

4. Auth & Multi-tenancy

• Add Firebase Auth or Auth0 for user accounts
• Namespace embeddings per organisation in the vector DB
• Role-based access for admin vs viewer

5. Observability

• Cloud Monitoring dashboards for latency / throughput
• Structured logging (JSON) shipped to Cloud Logging
• OpenTelemetry tracing for end-to-end request visibility

---

Tech Stack

Component	Technology
Backend API	FastAPI + Uvicorn
Frontend UI	Streamlit
Watermarking	Stegano (LSB steganography)
AI Features	MobileNetV2 (torchvision, pretrained)
Hash Similarity	imagehash (pHash)
Deep Similarity	numpy cosine similarity
Database	SQLite
Containerisation	Docker
Cloud Target	Google Cloud Run

---

License

MIT — free to use for hackathons, demos, and production.