This was a quick experiment I threw together to test the output of Azure Cognitive Services specifically with restaurant receipts.

• Python 3.6+ (using f strings, you can reformat and use with <3.6)
• Requests library
• Azure Subscription
  • Azure Form recognizer instance

When you create your Azure Form Recognizer instance, you'll get an endpoint and a key, you'll need both to make requests.

These values will need to be passed to the receipt-analysis.py file.

It's configured to use environment variables:

     REQUEST_URL = os.environ['REQUEST_URL']
     KEY = os.environ['REQUEST_KEY'] 

You'll need to set those variables up in your environment, or edit the receipt-analysis.py file and hardcode your values. A quick word on security: you really should avoid hardcoding, especially if your code might end up on a public code sharing platform.

receipt-analysis.py is what will make the request and takes the file path of your receipt picture as the first and only argument.

     python3 receipt-analysis.py [file-path]

If successful you'll see something like..

     Running OCR
     Loading Result...
     Waiting...
     Waiting...
     Results Loaded

It can take a few seconds to run as there are actually two requests, one of which can run multiple times before the results are available.

You should have a new file called 'azure-results.json' which will have the results from Azure.

You can use process-azure-results.py to print out certain information from the output of receipt-analysis.py. It'll print out the address and the items that were ordered.

     python3 process-azure-results.py [file-path-to-your-json]

I only did this as a POC so there is very little (i.e. no) error checking so depending on what you get back from Azure, you're likely to run into an error.