Data from Document Images

Your business requires users & merchants to fill various forms e.g. Loan Application, Merchant Registration, Account Activation, etc. These need identifying data (e.g. name, company name, address, ID documents: SSN [USA], Aadhaar ID [India]), and supporting documents (e.g. paycheck, bank statements, company registration). Users have to painstakingly enter details or your back-office personnel need to transcribe the information from scanned (copied) images of the documents. These approaches result in high support cost, user inconvenience, lower quality data, and low completion rates.

Autonom8 has built a rich Ingestion Toolkit to streamline this process. OCR delivers most of the text. We pick up where OCR leaves off, and use Computer Vision techniques & Machine Learning to retrieve the semantics: labels, attributes, values, headers, etc. These are then mapped to input elements in your forms.

We go beyond OCR and
retrieve semantic data from document images.
Semantic data retrieval

Consider the following paycheck fragment (click to magnify)

To a human reader the layout and tabular structure convey meaning i.e.:

  • "this person earned a Basic salary of Rs 48,667.00"

However, to the OCR system these are all just strings ("Earnings", "Amount", "(Rs)", "Basic", "48,667.00", etc.) and boundaries. Our Ingestion Toolkit extracts the meaning (semantics) from documents.

  • Basic Salary = INR 48,667.00

  • House Rent Allowance = INR 24,334.00

  • Cumulative Income Tax = INR 86,104.00

Combine the Ingestion Toolkit with the Form Bot Generator to build applications with integrated image processing capabilities.

  • Intelligence to understand:

    1. Document types: Driver's license, passport, ID documents: SSN [USA], Aadhaar, PAN [India], check leaf, paycheck, bank statement

    2. Document orientation (e.g. slanted), structure & layout (e.g. tabular)

    3. F&A vocabulary and entities: Name, address, company name, basic salary, etc.

  • Static Validation: Data ingestion can be coupled with rules to validate info e.g. start date must be earlier than end date, e-mail addresses must have a specific syntax

  • Dynamic Validation: Extracted data can be validated against external services e.g. does the Employee Name match the name on his Aadhaar card?

  • Web Service: All features of the Ingestion Toolkit are exposed as secure web services (REST).