🔍 Smart OCR - Multi-Engine Visual Text Recognition

Advanced browser-based OCR with multiple engines and visual results

Choose between fast mobile models or accurate server models. Features PPU-Paddle OCR, OnnxOCR, and Tesseract fallback. 100% client-side processing.

🚀 Try Live Demo | 📦 NPM Package | 📖 Documentation

✨ What's New in v4.0

🚀 Complete Multi-Engine Implementation

• PPU-Paddle-OCR Complete: Full implementation with deskew, detection, angle classification
• OnnxOCR TextSystem: Complete pipeline with TextDetector, TextRecognizer, TextClassifier
• Enhanced Processing: All preprocessing and postprocessing methods from both repositories
• Visual Results: Advanced visualization with polygon boxes and confidence scores

🎯 Advanced Features

• Auto-Deskew: Automatic image rotation correction
• Angle Classification: 180° rotation detection and correction
• Smart Padding: Adaptive padding for better recognition
• Batch Processing: Efficient batch recognition for multiple regions
• WebGL Acceleration: Optimized for mobile and desktop performance

🌟 Features

• 🔐 100% Private: All processing in browser, no data uploads
• ⚡ Multiple Engines:
  • PPU-Paddle-OCR: Fast mobile models with complete processing pipeline
  • OnnxOCR: Accurate server models with TextSystem architecture
  • Tesseract: Reliable fallback with word-level detection
• 🎨 Advanced Visualization:
  • Polygon and rectangle bounding boxes
  • Color-coded confidence scores
  • Reading order visualization
• 📱 PWA Support: Install as app, works offline with cached models
• 🌍 Multi-language: English, Chinese, Japanese, Korean with proper character dictionaries
• 💾 Smart Model Management:
  • Automatic model downloading with progress tracking
  • IndexedDB caching for instant switching
  • Model size optimization
• 📊 Comprehensive Metrics:
  • Per-stage processing time
  • Confidence scores for each detected text
  • Model performance comparison

🚀 Quick Start

Online Demo

Visit https://siva-sub.github.io/client-ocr-app

NPM Installation

npm install @siva-sub/client-ocr-app

Local Development

git clone https://github.com/siva-sub/client-ocr-app.git
cd client-ocr-app
npm install
npm run dev

📖 Documentation

Available OCR Engines

PPU Mobile (Fast) ⚡

• Optimized for mobile and real-time processing
• PP-OCRv5 detection + PP-OCRv4 recognition
• ~50-200ms processing time
• Best for: Live camera OCR, mobile apps

OnnxOCR v5 (Most Accurate) 🎯

• Latest PP-OCRv5 models with angle classification
• Highest accuracy on complex documents
• ~500-1000ms processing time
• Best for: Documents, receipts, forms

OnnxOCR v4 (Balanced) ⚖️

• Good balance of speed and accuracy
• PP-OCRv4 full pipeline
• ~300-500ms processing time
• Best for: General purpose OCR

OnnxOCR v2 (Server) 🖥️

• Heavy server models for maximum accuracy
• Designed for backend processing
• ~1000-2000ms processing time
• Best for: Batch processing, archives

Tesseract (Fallback) 🛡️

• Classic OCR engine
• Works offline without model downloads
• ~500-1500ms processing time
• Best for: Fallback, offline usage

Usage Guide

1. Select Engine: Choose based on your speed/accuracy needs
2. Upload Image: Drag & drop or click to select
3. Process: Models download automatically on first use
4. View Results:
   • Visual mode: See bounding boxes with confidence
   • Text mode: Get clean, copyable text
   • Download results as JSON

Model Management

• Models are cached after first download
• ✓ indicates cached models (instant loading)
• ⬇ indicates models need downloading
• Switch engines anytime without re-uploading

🛠️ Technology Stack

• PPU-Paddle-OCR: Mobile-optimized models
• OnnxOCR: High-accuracy ONNX models
• ONNX Runtime Web: Hardware-accelerated inference
• Tesseract.js: Classic OCR fallback
• OpenCV.js: Image preprocessing
• Vite + Mantine: Modern UI framework

📊 Performance Comparison

| Engine | Speed | Accuracy | Model Size | Use Case | |--------|-------|----------|------------|----------| | PPU Mobile | ⚡⚡⚡⚡⚡ | ⭐⭐⭐ | 12MB | Real-time OCR | | OnnxOCR v5 | ⚡⚡ | ⭐⭐⭐⭐⭐ | 25MB | Documents | | OnnxOCR v4 | ⚡⚡⚡ | ⭐⭐⭐⭐ | 20MB | General | | OnnxOCR v2 | ⚡ | ⭐⭐⭐⭐⭐ | 40MB | Batch | | Tesseract | ⚡⚡ | ⭐⭐⭐ | 11MB | Fallback |

🤝 Contributing

Contributions welcome! Please feel free to submit a Pull Request.

📝 License

MIT License - see LICENSE file

🙏 Acknowledgments

• PaddleOCR - State-of-the-art models
• PPU-Paddle-OCR - Mobile optimization
• OnnxOCR - ONNX implementation
• Tesseract.js - Classic OCR engine