gtfrans2re/docling-intercontinental-ocr

Comparative study of multilingual OCR (French, African, Hindi, and Indigenous Canadian languages) for RAG pipelines using Docling and Surya. # Fedora issue #123: Assignment submission document for the _processing multilingual documents_ phase 2 task. As an Outreachy contributor at the Fedora project, I completed an assigned task using Docling, an open-source optical character recognition (OCR) library, to convert scanned non-English documents (PDFs in my case) into machine-readable, editable digital text. Optionally, I completed the task using another OCR engine called Surya with the intent of analyzing and comparing its results with those of Docling through performance and output evaluation. ## OCR Performance on Non-Latin Scripts As an example, before getting started, this image demonstrates **Docling** processing the **Universal Declaration of Human Rights (UDHR)** in **Inuktitut (South Baffin)**. * **Bottom Left:** The terminal logs showing the `docling` command execution and the underlying **RapidOCR** engine initializing the necessary model files. * **Top:** The original source text written in **Canadian Aboriginal Syllabics**. * **Bottom Right:** The resulting **OCR Markdown output**, showing the tool's ability to capture the specific syllabic characters (e.g., ᐃᓄᒃᑎᑐᑦ) from the source. --- ## Repository Structure ``` . ├── assets/ │ ├── creating_ocr_virtual_env_with_python.png │ ├── docling_ocr_running_on_scanned_pdf_to_md.png │ ├── docling_ocr_version_output.png │ ├── docling_performance_output.png │ ├── loging_in_fedora_vm_with_multipasss.png │ ├── surya_ocr_installation.png │ ├── surya_ocr_running_on_scanned_pdf_to_markdown.png │ ├── surya_ocr_version_output.png │ └── tesseract_ocr_engine_version_output.png │ ├── data/ │ ├── amh.pdf │ ├── crm.pdf │ ├── frn.pdf │ ├── hnd.pdf │ ├── iku.pdf │ ├── swa.pdf │ ├── wol.pdf │ ├── yor.pdf │ └── zuu.pdf │ ├── output/ │ ├── docling/ │ │ ├── amh.md │ │ ├── crm.md │ │ ├── frn.md │ │ ├── hnd.md │ │ ├── iku.md │ │ ├── swa.md │ │ ├── wol.md │ │ ├── yor.md │ │ └── zuu.md │ …