techytushar / ocr-date-extractor

API to extract dates from documents using OCR

OCR Date Extractor

Flask API to extract dates from documents

How to use

The API is provided with 2 routes:

• If you want to pass Base64 encoded image, send a POST request with payload {"base_64_image_content": <base_64_image_bytes>} to

https://ocr-date-extractor.herokuapp.com/extract_date

• If you want to pass image file, send a POST request with payload {'image': <image_file>} to

https://ocr-date-extractor.herokuapp.com/extract_date_from_image

Python sample code to test out the API:

1. Sending the image as Base64 encoded

import requests, base64
img_url = <path_to_image>
with open(img_url, 'rb') as f:
    img = base64.b64encode(f.read())
response = requests.post('https://ocr-date-extractor.herokuapp.com/extract_date', data={'base_64_image_content':img})
print(response.content)

2. Directly uploading the file

import requests
url = "https://ocr-date-extractor.herokuapp.com/extract_date_from_image"
files=[
  ('image',('document.png',open('/Users/tushar/peak/document.png','rb'),'image/png'))
]
response = requests.post(url, data=payload, files=files)
print(response.text)

Working

The project performs the following steps for any given image:

• Re-scales the image if its too big in size
• Performs thresholding to separate foreground (the document) and the background
• Find contours and draws a bounding box on the document present in the image
• Crops the image to keep only the document
• Performs thresholding again to separate text from the background
• Apply OCR to extract text
• Use regex to extract out the date
• Date is then parsed and returned in YYYY-MM-DD format

Supported Date Formats

Following date format are supported with some flexibility:

• dd-mm-yyyy
• mm-dd-yyyy
• yyyy-mm-dd
• dd/mm/yyyy
• mm/dd/yyyy
• yyyy/mm/dd
• Aug23'19
• Feb 24, 2019
• 24 May'19

References

I took help from the following resources:

• Improving OCR Accuracy Medium
• OpenCV Docs
• Automatic Canny Edge PyImageSearch