4

Some companies provide software for Windows with their scanners* that can create PDFs from scanned pages which look exactly like the scanned material (as if it were just full-page images) but the text is recognized and copyable.

How can I create PDFs like this on Ubuntu?

Note that I don't want to convert scanned text into regular text. I would like to keep the resulting PDF looking picture-perfect as the original pages but add a recognized text layer over it for ease of use.

I have a working high-resolution scanner which I use with XSane currently. It scans the pages fine and creates beautiful, high-DPI images.

* namely, Canon with LiDE 220

Damn Terminal
  • 2,776

1 Answers1

5

Preamble

You are looking for a PDF sandwich, i.e. a scanned PDF with an invisible layer of text (or a layer of text which is simply placed behind the picture of each page).

There are several ways to create one. I will use the paper Term Weighting Approaches in Automatic Text Retrieval as an example of a document that needs OCR.

The pdfsandwich command

First of all, install this tool from the repositories:

sudo apt install pdfsandwich

Then you can just run it on your PDF file and wait:

pdfsandwich document.pdf

Screenshot of Evince showing a PDF sandwich

In the past, this method was not very precise, especially w.r.t. text positioning. It seems that now things got a lot better. Example from the PDF:

Abstract–The experimental evidence accumulated over the past 20 years indicates that

If you highlight the text in Evince, black boxes are shown.

PDF-XChange Viewer

This is a freeware, Windows-only program that works perfectly under Wine if you use the 32-bit version in a 32-bit Wine prefix. For this, I suggest using PlayOnLinux because it's very easy to select the latest Wine version and the fact that you want a 32-bit prefix.

Once installed, you can run it and select the OCR icon on the toolbar:

Screenshot of PDF-XChange Viewer under Wine

The output is usually very good and the text placement is precise. Example from the PDF:

Abstract--The experimental evidence accumulated over the past 20 years indicates that

If you highlight the text in Evince, the text is shown in a sans-serif font.

OCR.space

This is actually a web service. Go to ocr.space and select your file and language, then check the "Create searchable PDF with invisible text layer" option. Push the button and wait until the document is uploaded and converted.

Unfortunately, there is a bug for horizontal pages and they do not get rendered correctly in the output. I have notified the authors of this and they have acknowledged the problem.

OCRmyPDF

I am basing this part on what I posted in this answer on Super User.

OCRmyPDF is a multiplatform program written in Python, based on Ghostscript, Tesseract and Unpaper. From the docs:

What OCRmyPDF does

OCRmyPDF analyzes each page of a PDF to determine the colorspace and resolution (DPI) needed to capture all of the information on that page without losing content. It uses Ghostscript to rasterize the page, and then performs on OCR on the rasterized image to create an OCR “layer”. The layer is then grafted back onto the original PDF.

It can be easily installed on Debian and Ubuntu derivatives:

apt-get install ocrmypdf

You might need to install additional dictionaries for Tesseract, depending on the language you want to use. You can find a list of package names here (for macOS but they are the same on Ubuntu).

Usage is very simple and I suggest you use the optional -d (deskew) and -c (clean) parameters for better results. It will straighten every page and clean up small dots/imperfections before running the OCR process.

You can (and should) provide the language with -l.

Here's an example taken from this skewed document written in Italian:

Example for OCRmyPDF

The command I used was:

ocrmypdf -l ita -d -c input.pdf output.pdf

PDF24 hosts a free web-based version of OCRmyPDF that can be used without limitations.

Andrea Lazzarotto
  • 8,164