[FEATURE]: Make PDF Searchable: Overlay OCR'd text over Image?

[dcaud]

Input: A PDF
Output: A PDF

The output will look like the input, yet will have invisible OCR'd text in the same position as the text-as-image. I guess this could be accomplished using the bounding boxes from Google's JSON output and then using something like rMarkdown to recreate the PDFs (i.e., add the image to a page, then add the OCR'd text to the page)?

There is an example Python script here that essentially seems to do the job: https://shreevatsa.net/post/add-ocr-layer-to-pdf/

[Hegghammer]

Agreed. As a start, I reverse-engineered gcv2hocr, so I can convert .json files from Document AI to .hocr files, which in turn can be used to create PDFs with hocr-tools. It works fine for LTR languages, but I'm struggling with RTL scripts. Will publish when I have a robust solution.

[Hegghammer]

I have developed a temporary solution for this, described here: https://dair.info/articles/searchable_pdfs.

I will look more closely at the LaTeX approach when I have the time. The sticking point is whether TinyTex can handle it. If not, it would require users to have a working Latex installation on the side, and the solution wouldn't be R native any more.