OCR and digitization of plots

Optical character recognition

The tool of choice on Linux and MacOS is tesseract, an open source tool originally developed by HP.

Installation

On Linux (with apt package manager)

    sudo apt install imagemagick
    sudo apt install tesseract-ocr

and then install individual language packages

    apt-cache search --names-only tesseract-ocr
    sudo apt install tesseract-ocr-XXX (for language pack XXX, where XXX = swe, fra, ...)

For MacOS you find installation instructions here

Typical workflow

Convert image to tiff format and then use tesseract to convert to text. An extension .txt will automatically be appended to the output file.

So the following code will produce an intermediate file input.tiff and then convert to output.txt:

    convert input.png -resize 400% -type Grayscale input.tiff
    tesseract -l XXX input.tiff output  # convert using language XXX

Two examples below, both were available as jpg images:

English quote, light text on dark background

We need to reverse light and dark colours

    convert -negate -monochrome -density 300 english.jpg -depth 8 -strip -background white -alpha off english.tiff
    tesseract english.tiff english  # will add extension .txt, uses english as default language

which results in:

Continuing existence or cessation of
existence: those are the scenarios. Is it
more empowering mentally to work towards
an accommodation of the downsizings and
negative outcomes of adversarial
circumstance, or would it be a greater
enhancement of the bottom line to move
forwards to a challenge to our current
difficulties, and, by making a commitment
to opposition, to effect their demise?

Tom Burton
Long Words Bother Me

Swedish quote

Here we just convert to gray scale. However, if we don't specify the language (swe in this case) we get gibberish.

    convert -monochrome -density 300 swedish.jpg -type Grayscale swedish.tiff
    tesseract swedish.tiff swenglish

This produces

varit sé jfivla trfitt pa all!
53 j'zivla Hinge och nu bar
jag ingen jéivla aning om
hur jag kanner och vad
jag Sir trim p5.

whereas

    tesseract swedish.tiff -l swe swedish
    cat swedish.txt

produces

varit så jävla trött på allt
så jävla länge och nu har
jag ingen jävla aning om
hur jag känner och vad
jag är trött på.

PDF files

In order to convert pdf files, one needs (at least on Ubuntu) to change ImageMagick policy:

    sudo vi /etc/ImageMagick-6/policy.xml

and change <policy domain="coder" rights="none" pattern="PDF" /> to <policy domain="coder" rights="read/write" pattern="PDF" />

Note, that converting pdf documents to tiff will preserve all pages as different layers of that image, your image viewer may only show you the first page!

Image digitization

A very good tool with support for all kinds of plots is WebPlotDigitizer