DEV Community

Cover image for 📝✨ClearText
Ajinkya Bobade
Ajinkya Bobade

Posted on

4 4 2 3 3

📝✨ClearText

This is a submission for the GitHub Copilot Challenge : Transitions and Transformations

What I Built

I have built "ClearText" which is an AI-powered text detection and enhancement tool that makes text in images cleaner.

Title bar

It's Perfect For 🎯

  • 📄 Document Digitization
  • 📚 Book Scanning
  • 📱 Mobile Photos of Text
  • 🖨️ Improving Scanned Documents
  • 📑 Text Enhancement in Images

Demo

ClearText Demo

Repo

Github Repository - ClearText

Here's an example of what ClearText can do:

ClearText Demo

Image description

ClearText takes input image (left hand side), removes all noise and outputs pure text (right hand side).

ClearText has a huge potential where it can be used in the following fields:

Document Processing 📄

  • Banking & Finance
    • 🏦 Check processing
    • 📊 Financial statement digitization

Healthcare 🏥

  • Medical Records
    • 📋 Patient records digitization
    • 🔬 Lab report enhancement

Legal Industry ⚖️

  • Document Management
    • 📜 Contract digitization
    • 🗄️ Case file processing

Academic Use Cases 📚

  • 📖 Textbook scanning
  • 📑 Research paper digitization

Copilot Experience 🤖

I used co-pilot extensively to complete this amazing project. Here are the ways in which co-pilot helped me :

Code Completion 📝

  • Auto-completed common OpenCV operations
  • Suggested image processing parameters
  • Completed function signatures for Streamlit components

Chat Assistance 💬

  • Debugged ONNX model loading issues
  • Explained image processing pipeline
  • Suggested optimizations for image transformations

Inline Suggestions ⚡

  • Recommended error handling patterns
  • Suggested variable names and types

Model Switching 🔄

Used different models for specific tasks:

  • Code Completion: GitHub Copilot
  • Documentation: Claude
  • Debugging: GPT-4

Common Prompts Used 🎯

# Function implementation
/explain image processing pipeline
/suggest error handling
/optimize performance
Enter fullscreen mode Exit fullscreen mode

Code Edits ✏️

  • Refactored image processing functions
  • Added blur/no-blur options
  • Improved error messages
  • Enhanced documentation

Project Evolution & Contributions

Building on Open Source

This project builds upon the excellent CRAFT text detection model by CLOVA AI Research, while making significant architectural and functional improvements:

1. Production-Ready Architecture 🏗️

  • I converted the research-focused PyTorch model to production-ready ONNX format
  • Leveraged ONNX Runtime for optimized inference across different hardware
  • Added complete Docker containerization for reliable deployment

2. Enhanced Text Processing Pipeline 🔄

The original CRAFT model provides basic text detection. ClearText significantly expands on this by:

  • Adding custom image preprocessing for better text clarity
  • Implementing new post-processing transforms for enhanced output quality
  • Creating an entirely new text enhancement pipeline
  • Developing a user-friendly web interface for easy interaction

3. Major Output Improvements 📈

ClearText transforms the basic text detection output into a comprehensive text enhancement solution:

  • Original CRAFT: Basic text region detection
  • ClearText Additions:
    • Text clarity enhancement
    • Document digitization capabilities
    • Support for various document types (books, mobile photos, scanned documents)
    • Complete image processing pipeline

Transparency Statement

While this project builds upon CRAFT's foundational text detection capabilities, ClearText represents a significant evolution with entirely new functionality, architecture, and use cases. All original CRAFT code is properly credited and licensed under MIT License.

Conclusion

Developing ClearText during the GitHub Copilot 1-Day Build Challenge has been an amazing journey. Without co-pilot, transforming complex text detection model into an accessible, user-friendly web application would have been tremendously difficult. The project showcases how AI can bridge the gap between computer vision and practical, everyday use cases.

Retry later

Top comments (0)

Retry later
👋 Kindness is contagious

Please leave a ❤️ or a friendly comment on this post if you found it helpful!

Okay