OCR With Tesseract

Mix.install([
  {:kino, "~> 0.9"},
  {:deno_ex, "~> 0.2"}
])

Introduction

This Livebook shows you how you can use the Deno Terreract wrapper in order to perform optical character recognition (OCR for short) on images. Use the Kino input form to process your images and take it for a test drive!

Code

form = Kino.Control.form([file: Kino.Input.file("File")], submit: "Process image")

typescript_code = fn path ->
  {
    :stdin,
    """
    import { recognize } from "https://deno.land/x/tesseract/mod.ts";

    const output = await recognize("#{path}");
    console.log(output);
    """
  }
end

frame = Kino.Frame.new()

form
|> Kino.Control.stream()
|> Kino.listen(fn event ->
  path = Kino.Input.file_path(event.data.file.file_ref)

  image = path |> File.read!() |> Kino.Image.new(:png)

  {:ok, content} =
    path
    |> typescript_code.()
    |> DenoEx.run([], allow_run: true)

  ocr_result = String.trim(content) |> Kino.Markdown.new()

  render_result = Kino.Layout.grid([image, ocr_result], columns: 2)
  Kino.Frame.render(frame, render_result)
end)

frame