Reading image text using Elixir

This post will detail how you can read text off of an image using the Elixir programming language.

To handle this we will use a technology called Optical Character Recognition which is used to find printed or handwritten text characters inside of an image.

The steps we will take to complete this project are:

  1. Install the system package for tesseract (an OCR engine).
  2. Include the tesseract-ocr-elixir lib in your elixir dependencies.
  3. Test the library functionality using IEx.

Important note before installation:

If you don't install the tesseract engine before trying to use the Elixir library tesseract-ocr-elixir you will likely be met with this error: 
iex(1)> TesseractOcr.read(".lib/testocr.png")

** (ErlangError) Erlang error: :enoent

    (elixir 1.12.0) lib/system.ex:1041: System.cmd("tesseract", [".lib/testocr.png", "stdout"], [])

    (tesseract_ocr 0.1.5) lib/tesseract_ocr.ex:19: TesseractOcr.read/2

This error is letting you know you haven’t installed tesseract, not that the file path is empty (which without reading the stacktrace you might suspect because of the :enoent eror).

Once you install tesseract this error should go away. You can see the system request to tesseract in the screenshot below:

replace_image_0_x


1. Installing Tesseract

I use Homebrew to install system dependencies on my mac. So for me installing this library was straightforward:

brew install tesseract

The tesseract website has more options for installation.

2. Add tesseract-ocr-elixir lib to deps

In your mix application add the library tesseract-ocr-elixir to deps. This is an Elixir wrapper for OCR.

def deps do
  [
    {:tesseract_ocr, "~> 0.1.5"}
  ]
end

Run mix deps.get to install the package.

3. Test the library functionality

To do a quick test of the library you will need an image with text, so grab your favorite meme and then load your application using iex -S mix.

Now you can test the library by the read function which will print out any words OCR finds:

iex> TesseractOcr.read("test/resources/testocr.png")
"OH YOU FOUND SOME INTERNET MEMES YOU MUST BE FUNNY"

That was the last step! You can also use the library to read PDFs and do other fun things.