Using Tesseract on Heroku with Django

Question

I would like to add OCR capabilities to my Django app running on Heroku. I suspect the easiest way is by using Tesseract. I've noticed that there are a number of python wrappers for Tesseract's API, but what is the best way to get Tesseract installed and running on Heroku? Via a custom buildpack like heroku-buildpack-tesseract maybe?

Answer 1

Heroku, Django and tesseract

This doc will walk you through setting up tesseract on Heroku (i'm using django)

Steps

1) Add heroku-apt-buildpack using the command:

This is the stable version. See the source repository

$ heroku buildpacks:add --index 1 heroku-community/apt

2) Add Aptfile to project directory

`

$ touch Aptfile

3) Add the folowing to the Aptfile

tesseract-ocr-eng is the English language file for tesseract.

tesseract-ocr
tesseract-ocr-eng

4) Get path to the data downloaded by the tesseract-ocr-eng package

We will use this path for the next step

$ heroku run bash
$ find -iname tessdata # this will give us the path we need

You can exit heroku shell now exit

5) Now set a heroku config variable named TESSDATA_PREFIX to path

Set a heroku config variable named TESSDATA_PREFIX to the path returned from find -iname tessdata cmnd above

$ heroku config:set TESSDATA_PREFIX=./.apt/usr/share/tesseract-ocr/4.00/tessdata

Now set heroku set a heroku config variable named TESSDATA_PREFIX to the path returned from find -iname tessdata

6) Push changes to heroku

Set a heroku config variable named TESSDATA_PREFIX to the path returned from find -iname tessdata cmnd above

$ git push heroku master

I hope this helps. Let me know if it works for you.

Answer 2

I'll try to capture some notes on the solution I arrived at here.

My .buildpacks file:

https://github.com/heroku/heroku-buildpack-python
https://github.com/clearideas/heroku-buildpack-ghostscript
https://github.com/marcolinux/heroku-buildpack-libraries

My .buildpacks_bin_download file:

tesseract-ocr https://s3.amazonaws.com/tesseract-ocr/heroku/tesseract-ocr-3.02.02.tar.gz 3.02 eng,spa

Here is the key piece of python that does the OCRing of pdf files:

        # Additional processing
        document_path = Path(str(document.attachment_file))

        if document_path.ext == '.pdf':
            working_path = Path('temp', document.directory)
            working_path.mkdir(parents=True)

            input_path = Path(working_path, name)
            input_path.write_file(document.attachment_file.read(), 'w')

            rb = ReadBot()

            args = [
                'VBEZ',
                # '-sDEVICE=tiffg4',
                '-sDEVICE=pnggray',
                '-dNOPAUSE',
                '-r600x600',
                '-sOutputFile=' + str(working_path) + '/page-%00d.png',
                str(input_path)
            ]

            ghostscript.Ghostscript(*args)
            image_paths = working_path.listdir(pattern='*.png')
            txt = ''

            for image_path in image_paths:
                ocrtext = rb.interpret(str(image_path))
                txt = txt + ocrtext

            document.notes = txt
            document.save()
            working_path.rmtree()