Classification and Data extraction from Identity documents

Executive Summary

A UK-based client working with various government agencies faced critical challenges in verifying identity documents submitted by citizens for public services. The manual document verification process was time-consuming, error-prone, and lacked scalability. To solve this, we implemented an AI-powered solution that automates the extraction and classification of information from identity documents, including verifying handwritten SRA numbers provided by solicitors. By integrating Google Vision API and advanced AI algorithms, the client now benefits from improved accuracy, operational efficiency, and enhanced fraud detection.

Problems faced by Client

The client was responsible for verifying identity documents submitted by the public to access essential government services. Their existing process was entirely manual, leading to several operational challenges:
•Human Errors: Manual data entry and document verification were prone to mistakes.
•Time-Consuming Processes: High volume of documents led to delays in service delivery.
•Inconsistency: Varying document formats and handwritten annotations reduced reliability.
•Fraud Risk: No standardized way to verify solicitor-attested documents, increasing the risk of forged submissions.


Proposed Solutions

To address these issues, we developed an AI-based document verification system with the following capabilities:
•Document Classification: Automatically identify and categorize document types (e.g., passport, driving license, council tax bill).
•OCR & Handwriting Recognition: Extract text, including handwritten SRA numbers, from scanned document images.
•Data Verification: Cross-check extracted SRA numbers against third-party databases such as the Solicitors Regulation Authority (SRA) registry.
•Compliance Management: Ensure GDPR compliance through appropriate use of cloud services and data residency options.

How It Works

1.Image Input: The user uploads scanned identity documents into the system.
2.Data Extraction: Google Vision API processes the image, applying Optical Character Recognition (OCR) and AI-based classification.
3.SRA Number Detection: AI algorithms detect handwritten SRA numbers where applicable.
4.Verification: The extracted SRA number is validated against an authorized solicitor verification portal to confirm authenticity.
5.Output Generation: The system compiles extracted data including:
First and last name
Date of birth
Address
Document identification number
Document type classification
6.Final Validation: Verified and structured data is forwarded to the government service portal for processing.

Technologies Used

•Google Cloud Vision API : For advanced OCR and document structure recognition.
•AI/ML Algorithms : For document classification and handwritten text recognition.
•Cloud Services : For scalable and secure processing with data residency compliance.
•Third-party API Integration : A third-party SRA identity verification website used this AI model.
•Coding Language: Dot Net/HTML.

Client Benefits

•Increased Accuracy: Significantly reduced errors in data extraction and document validation.
•Faster Processing: Cut down verification time from hours to minutes, enabling quicker service delivery.
•Fraud Prevention: Reliable SRA verification ensures solicitor legitimacy and prevents document forgery.
•Operational Efficiency: Automated workflows reduce dependency on manual staff and enable higher throughput.
•Regulatory Compliance: Ensures GDPR compliance with careful handling of personal and sensitive data.
•Scalability: Easily accommodates growing document volumes without compromising performance.

Sample Demo Images to display the Extraction Output

Photocopy image with SRA Number

Details extracted – “Extracted Text: UK DRIVING LICENCE 1. FRITH 2. ASHLEIGH JAMES ERNEST 3. 02.03.1987 UNITED KINGDOM 4a.28.05.2021 4c. DVLA 4b.27.05.2031 5. 7 FRITH803027AJ9EN 97 FRITH 23 REDLAND WAY, AYLESBURY HP21 9RJ 9. AM/A/B/f/k/q 米 Atique Mazher SRA number: 51873 Labels: Identity document (95.85%), Paper Product (77.34%), License (60.42%), Document (58.70%) Colors: RGB(91,84,77) – 25.27%, RGB(192,188,185) – 23.62%, RGB(163,158,152) – 19.13%, RGB(62,54,49) – 15.04%, RGB(120,114,107) – 13.28% Landmarks: No landmarks detected Logos: Union Jack (78.18%) Safe Search: Adult: VERY_UNLIKELY, Violence: VERY_UNLIKELY, Racy: VERY_UNLIKELY”