nayyhah / PDFAutomation-OCRTextRecognition

PDF Automation - OCR Text Recognition

DECIPHER 2.0 | PDF Automation - OCR Text Recognition

Problem Statement

There are various validation rule engines across the industry. There are all Converting PDF documents into structured data which you can use in other systems is however not a trivial task and many businesses find themselves manually re-keying product data from PDF documents into spreadsheets or databases.

Extraction of critical data from those PDF’s(Invoice, receipts, sales orders) and in return receive structured data. Need to convert PDF to text for download

Assumptions/Callouts

• Any data set can be used. Use supply chain related data
• Pick few sample PDF's with watermark

Solution - DECIPHER 2.0

Decipher 2.0 is a user-friendly tool that identifies the data items one wants to extract from their uploaded invoice and gives back the desired fields as a downloadable file in multiple formats.It comes with the following built-in functionalities:

Extracts useful information from large, multiple PDFs in just seconds

Adjusts the alignment of randomly oriented pdf images

Remove watermarks to provide flawless results

Store the extracted information in cloud storage for future access

Architecture

Technology Stack

Frontend

• React JS : web-app user interface
• Netlify : deploying the web-app

Backend

• Express.js : setting up backend server
• Heroku : deploying backend services

Storage

• MongoDB : storing schemas
• Cloudinary : cloud storage for files

Tools and Languages

• Tesseract : extracting text through OCR
• Python : parsing pdf and images

Screenshots

Future Additions

Bounding Box Flexibility : Users can select their own bounding boxes and this feature can be implemented using JCrop Library

Language Translation : PDFs can be translated from one language to other using libraries like textblob or googletrans

QR Code Reader : Values of QR Codes or Bar Codes in PDFs can also be decoded and then stored using pyzbar library

Processing Handwritten Invoices : Text from handwritten invoices can be extracted and parsed. For this, Tensorflow library can be used to create and train a neural network by using datasets of extracted invoices.