Overview

SmolDocling-OCR provides a streamlined workflow for converting scanned documents and images into machine-readable text. The system leverages Tesseract for OCR, OpenCV for image pre-processing, and a simple web interface for uploading images and viewing results. Designed for researchers and students, it emphasizes accuracy, speed, and ease of deployment.

Tech Stack

Challenges

• Handling diverse document qualities and layouts (skew, noise, handwriting).
• Optimizing pre-processing to improve OCR accuracy across languages.
• Building a simple, user-friendly web interface for non-technical users.
• Ensuring lightweight deployment suitable for resource-constrained environments.

Solution

The pipeline integrates image enhancement with OpenCV and Pillow, followed by text extraction using Tesseract OCR. A Flask backend processes uploads and OCR tasks, while a React frontend allows users to submit documents and view extracted text. The system is containerized for easy deployment.

Outcome

SmolDocling-OCR enables rapid and reliable text extraction from various educational and research documents. Its lightweight design allows deployment on low-resource hardware, and its modular structure supports easy customization for different research needs.