README.md

metadata

title: YouTube Slide Extractor
emoji: 🎓
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 4.20.0
app_file: app.py
pinned: false
license: mit

🎓 YouTube Slide Extractor

An AI-powered tool that automatically extracts slides from educational YouTube videos using computer vision and OCR technology.

🚀 Features

• Automatic Slide Detection: Uses computer vision to identify unique slides in videos
• OCR Text Extraction: Extracts text content from slides using Tesseract OCR
• PDF Generation: Creates downloadable PDF documents from extracted slides
• Smart Filtering: Removes duplicate frames and focuses on content changes
• Batch Download: Download all results in a convenient ZIP file

🎯 How It Works

1. Video Download: Downloads the YouTube video using yt-dlp
2. Frame Extraction: Extracts frames at regular intervals (every 5 seconds)
3. Slide Detection: Uses structural similarity to identify unique slides
4. OCR Processing: Extracts text content from each slide
5. PDF Creation: Generates a PDF with slides and extracted text
6. Results Packaging: Creates a ZIP file with all slides, PDF, and metadata

📋 Usage Instructions

1. Enter Video URL: Paste a YouTube video URL in the input field
2. Configure Options: Choose whether to generate a PDF (recommended)
3. Start Extraction: Click the "Start Extraction" button
4. Monitor Progress: Check the status updates as processing occurs
5. Download Results: Download the ZIP file and/or PDF when complete

🔧 Technical Details

Computer Vision Pipeline

• Frame Extraction: OpenCV for video processing
• Similarity Detection: Structural Similarity Index (SSIM) for duplicate detection
• Image Processing: Preprocessing for improved OCR accuracy

OCR Technology

• Engine: Tesseract OCR with optimized configurations
• Preprocessing: Noise reduction and contrast enhancement
• Text Extraction: Configurable OCR parameters for best results

Supported Video Types

• Educational lectures and presentations
• Tutorial videos with slide content
• Webinars and online courses
• Conference presentations

⚠️ Limitations

• Video Length: Longer videos may take more time to process
• Quality Dependency: OCR accuracy depends on slide image quality
• Text-Heavy Content: Works best with slides containing clear text
• Processing Time: Typically 1-3 minutes for a 10-minute video

🛠️ Technology Stack

• Frontend: Gradio for the web interface
• Computer Vision: OpenCV and scikit-image
• OCR: Tesseract via pytesseract
• Video Processing: yt-dlp for YouTube downloads
• PDF Generation: ReportLab for document creation
• Image Processing: PIL/Pillow for image manipulation

📊 Performance Tips

• Shorter Videos: Process videos under 30 minutes for faster results
• High Quality: Use videos with clear, readable slides
• Stable Content: Works best with static presentation slides
• Good Lighting: Videos with good contrast produce better OCR results

🔒 Privacy & Security

• No Data Storage: Videos and slides are processed temporarily
• Local Processing: All processing happens on the server instance
• Automatic Cleanup: Temporary files are cleaned up after processing
• No User Tracking: No personal data is collected or stored

📄 License

MIT License - Feel free to use and modify for your projects.

---

Note: This tool is designed for educational purposes and respects YouTube's terms of service. Please ensure you have permission to download and process videos.