[Bug]: The file size increases significantly by OCR even without image recompression about ocrmypdf

[ybeltukov]

It seems to me that it is related to hocr pdf renderer, which is enabled by default now. It produces a better visual quality (see e.g. #1131 ), however it increases the size of OCR layer almost twice.

With the option --pdf-renderer sandwich I obtain the following sizes for the same file:
251 KB → 310 KB → 369 KB

So the OCR layer takes 59 KB for sandwich and 98 KB for hocr

So the questions are:

• Is it possible to optimize hocr renderer?
• Is it possible to remove previously added OCR layer without image recompression?
  (--force-ocr is not suitable for this task)

[Jmuccigr]

I wrote a script that redoes the ocr on PDFs by deleting any original text from the file and then using ocrmypdf to generate new ocr which I then add to the original file. I use it mainly to replace the often bad ocr in jstor files. It uses Ghostscript to remove the text and relies on some other stuff that you can see in the code.