COMPUTER VISION
Universal OCR System for ID Documents: Smart Acquisition & Recognition
I developed a robust OCR system designed for reading identity documents of varying formats and origins. The solution is capable of extracting data from a wide range of IDs—cards and passports—adapting dynamically to the diversity of document layouts, languages, and security features.
The entire workflow was engineered for seamless integration with mobile applications, supporting a fast and reliable digital onboarding process.
This OCR solution is currently used by SNAI and is fully operational in their retail stores across Italy, supporting real-world customer identification workflows.
Project Goals & Approach
The main challenge was to create an OCR engine flexible enough to handle multiple document types from different countries, each with its own format and features.
- Implemented a generalizable method for detecting and parsing data, making the system scalable to new documents with minimal adjustment.
- Trained and tuned OCR models to manage multiple languages, fonts, and anti-fraud patterns found on real IDs and passports.
- Ensured the architecture is modular and ready for fast adaptation to any future document type.
Server-Side OCR Service & Integration
The OCR logic runs on a dedicated server, exposing APIs designed to interact efficiently with external applications.
- Deployed the backend OCR service on a secure server for performance and scalability.
- Provided endpoints that receive document images, process them, and return structured identity data.
- Integrated the system with an Android app built using Ionic, allowing mobile users to scan and verify documents in real time.
Camera Intelligence: Border, Focus, and Sharpness Detection
To maximize recognition quality and user experience, I developed advanced logic for document acquisition on mobile.
- Engineered an algorithm for live detection of document borders, focus, and image sharpness before capture.
- Implemented a smart auto-capture: as soon as the document is fully visible, properly aligned, and sharp, the camera triggers automatically.
- This minimizes user error and ensures only optimal images are processed for OCR, resulting in higher accuracy and smoother onboarding.
Impact & Next Steps
The universal OCR and smart capture system has significantly reduced friction in digital onboarding and KYC (Know Your Customer) flows.
- Achieved robust, automated extraction of personal data from global identity documents.
- Reduced manual review and sped up user verification times for client applications.
- The solution is currently powering SNAI’s identification workflows in retail locations, enabling automated customer onboarding at scale.
- Future work includes expanding support for new document classes and enhancing anti-fraud checks using AI.