Optical Character Recognition

Incode’s in-house developed Optical Character Recognition (OCR) technology extracts and processes data from thousands of global identity document types, ensuring fewer data discrepancies and enhanced fraud detection.

Industry leaders trust Incode with their AI fraud prevention

Key challenges

Why we invested into our OCR toolkit

Many general-purpose OCR technologies on the market fail to meet the requirements of fast-moving businesses.

Multiple document types

Basic OCR cannot handle diverse documents worldwide because it lacks ML models trained on large, varied ID datasets.

Text readability & font differences

Fonts with various shapes and styles often confuse basic OCR, leading to errors unless the system is trained to handle them.

Language limitations & special characters

Handling multiple languages challenges OCR, especially when scripts use many characters, diacritical marks, or different reading directions.

Reading tricky symbols

Special symbols, like those on US bank checks, are often missed by general OCR systems that are not trained to detect them.

Confusing designs & document similarities

OCR struggles with poor contrast, cluttered layouts, and overlapping objects. It can also misread similar documents such as permits and licenses.

Dependency on third-party developers

OCR that relies on third parties adapts slowly, often misreading government documents with new elements the models do not recognize.

Challenging environmental conditions

Poor lighting, shadows, stains, or tears can reduce even advanced OCR accuracy and reliability.

Poor image quality

Basic OCR struggles with blurry or distorted images, often misreading characters and shapes.

Our solution

The benefits of Incode OCR technology

Built in-house, Incode OCR processes 4.9K+ identity documents worldwide. It adapts to low-quality images, complex fonts, and unique symbols with accuracy and scale.

Purpose-built for global IDs

Unlike general tools, Incode OCR is designed to capture and process data from thousands of identity documents worldwide.

Recognizes complex elements

We use advanced ML models that adapt to document variations like fonts, diacritics, symbols, and barcodes for stronger OCR results.

Built for global scalability

Our in-house technology uses a 2-stage algorithm to accurately extract text from 4.9K+ documents in 190+ countries.

Ensures regulatory compliance

We improve accuracy so your organization meets regulations and protects itself from costly penalties or reputational loss.

Top speed operation

Our ML-driven capture SDK reviews and validates multiple frames within seconds, surpassing human performance.

Real-time feedback & optimization

Incode’s capture SDK improves image quality through smart frame selection, real-time feedback, and automatic ID orientation detection.

Full air gap solution

Our full air gap solution enhances security in special use cases, protecting against unauthorized access, data leaks, and cyberattacks.

Stay one step ahead

We design pipelines for specific document types, enabling fast adaptation to new formats and updates.

Experience AI-powered OCR excellence

Trial error-free verification with Incode’s intelligent technology.

How it works

Our OCR technology toolkit

This guide showed how Incode’s in-house developed OCR uses ML to improve the accuracy of identity verification.

ID Capture

ID Classification

ID OCR

Barcode Reader

Entities extraction & representation

ID Capture

SDK processes a videostream of the document, auto-detects orientation, and reads NFC chips when available.
ML models evaluate frame quality in milliseconds.
Real-time feedback helps users fix lighting, focus, or visibility issues.
A final ML check confirms a clear frame before capture is complete.
If capture fails, the user is prompted to take a photo, which is then validated.

ID Classification

Candidate proposal

Extracts document features such as layout, text placement, fields, and colors.
Converts them into a numerical vector.
Compares the vector with a database of known IDs.
Generates candidate document types (e.g., passport, driver’s license).

Refinement

Uses text analysis to detect keywords or symbols (e.g., “permanent” vs “temporary”)
Runs field-specific checks on numbers, authorities, and dates
Selects the final document type.

ID OCR

The system detects where words are located on the document, using document-type data to focus on important fields. It defines word boundaries to handle even dense text.
It then reads and interprets the words, guided by field formatting from the classification stage, which improves accuracy for structured fields and special symbols.

Barcode Reader

Barcodes hold key user data but are difficult for OCR to read.
Our ML model restores and enhances low-quality images to ensure accurate reading.

Entities extraction & representation

We accurately extract key entities such as names, addresses, and document numbers.
The system structures this data while adapting to date formats, field shifts, and document-specific rules like front or back address stickers.

User experience

Our streamlined workflows

Our ML models handle all the heavy lifting, making every user interaction feel effortless. By delivering improved results, even in suboptimal conditions, we save our users time and minimize the need for manual intervention.

Features such as smart frame selection, automatic ID orientation detection, and real-time feedback help to ensure our process is straightforward and easy to navigate. By simplifying and speeding up the process, we boost completion rates and drive conversions.

Highest accuracy

Proven results against leading OCR providers

Incode OCR outperforms open-source and general purpose solutions. With ongoing benchmarking and bias analysis, we ensure consistent accuracy across document types, languages, and visual conditions.

Performance analysis

Our OCR delivers higher accuracy across diverse document types, outperforming general-purpose solutions in key fields such as name, address, birthplace, document number, expiry date, and machine-readable zones (MRZ).

Benchmark comparison

In-house testing compared Incode’s OCR with Google’s general-purpose OCR. We measured exact matches across key fields from common document types. Incode consistently achieved higher accuracy.

Bias analysis

We regularly evaluate model performance across document types, languages, and visual conditions to reduce bias in recognition and ensure consistent accuracy.