README ocr-tesseract-textract

Goal

This code shows how easy it is to use AWS Textract to synchronously OCR a single page. By "synchonous", we mean we submit a page to Textract and wait for the result.

Operation

Textract expects an image format, JPG or PNG. Scanned PDF docs have the image embedded in JPG format, so we first extract that binary data then send it to Textract. We parse the BLOCKs and LINEs and build the text output. This is then printed.

Creating a "scanned" image

To get a fake "scan" PDF, I created a one-page doc of text, converted it to a JPG, then converted that to a PDF.

Since our original text is something we know, we should be able to compare the accuracy of our extracted text with the original to determine accuracy. That's for later.

Invoke

Invoke like:

./textract.py palefire.jpg.pdf

Developer environment setup

I used pyenv to select python version 3.7.2 then created and activated a virtual environment, then installed boto3:

pyenv local 3.7.2
python -mvenv .venv
pip install -r requirements.txt