Form Recognizer with Azure Functions

Pre-Requisites

• Python 3
  • Virtual Python Environment Builder
• Visual Studio Code
  • Python Extension for Visual Studio Code
  • Azure Account Extension for Visual Studio Code
  • Azure Functions Extension for Visual Studio Code
• Docker Desktop
• Postman
• Azure Storage Account
• Form Recognizer Resource
• Trained Form Recognizer Model

Environment Setup

1. Azure Portal > Create a Function App

   • App name: Provide your Function App a unique name
   • Resource Group: Create a new or select an existing Resource Group
   • Operating System: Linux
   • Hosting Plan: Consumption
   • Location: Select a Location (e.g. West Europe)
   • Runtime Stack: Python

2. Visual Studio Code > Create a New Virtual Envrionment (Python).

   • Open Visual Studio Code

   • Open the folder that you would like to work from

   • Terminal: virtualenv . (this will create a virtual environment)

   • Terminal: .\Scripts\activate (this will activate the virtual environment)

     

3. Press: Ctrl + SHIFT + P

4. Search "Create New Project", Select "Azure Functions: Create New Project"

5. Select the folder to deploy the Azure Function project to.

6. Select Python as the language for your Azure Function project.

7. Select HTTP trigger as the template for your projects function.

8. Provide the function a name (e.g. AnalyzeForm).

9. Select the Function authorization level.

Python Library Requirements

• Run the following command in your activated environment via Terminal: pip install azure-storage-blob
  
• Add the following line to requirements.txt: azure-storage-blob==1.4.0
  

Python Code Sample

This can be copy and pasted into __init__.py

import requests
import logging
import azure.functions as func
from azure.storage.blob import BlockBlobService

SUPPORTED_CONTENT_TYPES = ['application/pdf', 'image/jpeg', 'image/png']

def main(req: func.HttpRequest) -> func.HttpResponse:
    # 1. Get Header Values - Azure Blob Storage
    storage_account_name = req.headers.get('storage_account_name')
    storage_account_key = req.headers.get('storage_account_key')
    container_name = req.headers.get('container_name')
    blob_name = req.headers.get('blob_name')

    # 2. Get Header Values - Form Recognizer Cognitive Service
    region = req.headers.get('region')
    model_id = req.headers.get('model_id')
    subscription_key = req.headers.get('subscription_key')
    
    # 3. Get Form from Blob Storage
    blob_service = BlockBlobService(account_name=storage_account_name, account_key=storage_account_key)
    blob = blob_service.get_blob_to_bytes(container_name, blob_name)
    data = blob.content
    content_type = blob.properties.content_settings.content_type

    body = None
    if content_type in SUPPORTED_CONTENT_TYPES:

        # 4. Generate Endpoint and HTTP Header
        endpoint = 'https://{0}.api.cognitive.microsoft.com/formrecognizer/v1.0-preview/custom/models/{1}/analyze'.format(region, model_id)
        headers = {
            'Content-Type': content_type,
            'Ocp-Apim-Subscription-Key': subscription_key
        }

        # 5. Analyze Form
        response = requests.post(endpoint, headers=headers, data=data)
        body = response.content

    # 6. Return HTTP Response
    return func.HttpResponse(body,headers={'Content-Type':'application/json'})

Test the Azure Function Locally using Postman

1. Execute the following command in Terminal to launch the functions runtime host: func start

2. Open Postman and create a new request with the following properties:

   • Method: GET
   • URL: http://localhost:7071/api/YOUR_FUNCTION_NAME

3. Add the following Headers:

   Key	Value
   storage_account_name	STORAGE_ACCOUNT_NAME
   storage_account_key	STORAGE_ACCOUNT_KEY
   container_name	STORAGE_CONTAINER_NAME
   blob_name	STORAGE_BLOB_NAME
   region	FORM_RECOGNIZER_RESOURCE_REGION
   model_id	FORM_RECOGNIZER_MODEL_ID
   subscription_key	FORM_RECOGNIZER_SUBSCRIPTION_KEY

4. Click Send

If successful, this should return the results of the form being processed by the Analyze Form Recognizer API.

Publish the Azure Function

Execute the following terminal command to publish the current directory contents to an Azure Function App:
func azure functionapp publish NAME_OF_YOUR_FUNCTION_APP --build-native-deps

Check out https://aka.ms/func-python-publish for more information on publishing a Python function to Azure.

Test the Azure Function

Once the deployment has completed successfully, a public invoke url will be returned to the terminal window.

1. Copy the invoke url to the clipboard.

2. Return to Postman and duplicate the existing request by right-clicking on the tab and clicking Duplicate Tab.

3. Replace the http://localhost/api/YOUR_FUNCTION_NAME with the invoke url.

4. Click Send

   

That's it! If successful, this should return the same results as before but this time, the function will be running in the cloud. We now have a serverless method of processing forms stored on Azure Blob Storage using Azure Functions.