📄 DocScan Gallery Benchmark

Document Scanner Quality Benchmarking Platform

10-metric automated quality pipeline for document scanning apps

📖 Metrics · 🚀 Quick Start · 📊 Reports

---

📌 What Is This?

A Python-based benchmark platform that automatically evaluates the output quality of document scanning applications (Samsung Notes Scanner, Microsoft Lens, Google PhotoScan, Adobe Scan). It runs 10 quality metrics on scanned document images and generates a comparative report.

Used to benchmark document scanner quality across 3 apps on 50 real-world documents.

---

🎯 Why This Exists

Document scanners are judged subjectively — "this looks better." This tool makes quality measurable and comparable using computer vision metrics, giving QA teams objective pass/fail criteria.

---

📐 Quality Metrics (10)

#	Metric	What It Measures
1	OCR Accuracy	Text recognition via Tesseract — character error rate
2	SSIM	Structural similarity vs ground truth scan
3	PSNR	Signal-to-noise ratio — image fidelity
4	Perspective Correction	Geometric distortion after de-warping
5	Shadow Removal	Shadow detection and elimination quality
6	Contrast Enhancement	Text readability improvement score
7	Edge Sharpness	Laplacian variance — text edge clarity
8	Background Whiteness	Page background normalization quality
9	Noise Level	High-frequency artifact detection
10	Color Accuracy	ΔE2000 vs reference document colors

---

🏗️ Architecture

doc-scan-gallery/
│
├── app/
│   ├── core/
│   │   ├── benchmark_engine.py    ← Orchestrates all metrics
│   │   ├── ocr_engine.py          ← Tesseract OCR integration
│   │   ├── visual_validator.py    ← SSIM, PSNR, edge metrics
│   │   ├── scoring_engine.py      ← Weighted score calculation
│   │   ├── semantic_validator.py  ← Content preservation check
│   │   └── report_generator.py   ← HTML + PDF report output
│   │
│   ├── database/
│   │   └── db_manager.py          ← SQLite results storage
│   │
│   ├── models/
│   │   └── data_models.py         ← ScanResult, BenchmarkReport
│   │
│   └── resources/
│       └── styles/
│           └── dark_theme.qss     ← UI theme
│
├── test_documents/                ← 50 ground truth documents
├── scan_inputs/                   ← App-scanned versions
├── reports/                       ← Generated benchmark reports
├── requirements.txt
└── README.md

---

🚀 Quick Start

# Clone
git clone https://github.com/vanichalla24/doc-scan-gallery-.git
cd doc-scan-gallery-

# Install
pip install -r requirements.txt

# Add test documents
cp your_documents/* test_documents/
cp your_scans/* scan_inputs/

# Run benchmark
python -m app.core.benchmark_engine --input scan_inputs/ --output reports/

# View report
open reports/benchmark_report.html

---

📊 Output Report

BENCHMARK RESULTS — Samsung Notes Scanner vs Microsoft Lens
────────────────────────────────────────────────────────────
Metric              Samsung Notes    Microsoft Lens   Winner
────────────────────────────────────────────────────────────
OCR Accuracy        94.2%            91.8%            Samsung ✅
SSIM Score          0.891            0.867            Samsung ✅
Perspective Corr.   97.1%            95.4%            Samsung ✅
Shadow Removal      88.3%            92.1%            Lens    ✅
Edge Sharpness      82.4             79.1             Samsung ✅
────────────────────────────────────────────────────────────
OVERALL SCORE       91.4/100         89.2/100         Samsung ✅

---

🔧 Requirements

opencv-python>=4.8.0
pytesseract>=0.3.10
scikit-image>=0.21.0
numpy>=1.24.0
Pillow>=10.0.0
reportlab>=4.0.0
sqlite3 (built-in)

---

👩‍💻 Author

Vaani Challa — QA Architect | Samsung SRIB | 17+ Years DocScannerBench contributor · Innovation Award winner · Galaxy S25 + S26 Best Award

---

📄 License

MIT © Vaani Challa