PingOne Verify and documentation authentication

Document authentication in PingOne Verify uses machine learning and multiple verification checks to assess the validity of a document presented by a user.

What is machine learning and how is it used in document authentication?

Machine learning (ML) is a branch of artificial intelligence that allows computers to recognize patterns and make predictions based on data. In the context of document authentication (DocAuth), ML is employed to assess whether a document is genuine or fraudulent by comparing the features of the submitted document to a trained model.

Machine learning helps detect:

Printed or photocopied documents
Document tampering, such as changes to security features like holograms, text, or watermarks
Subtle forgeries, such as improper alignment of text or altered photos

Models are trained on thousands of document samples to identify genuine versus fake documents, allowing for more accurate detection over time.

This continual learning process improves the accuracy of the verification system, particularly in high-volume regions where more document types are processed.

What are the steps for Automated Document Authentication?

The document authentication process in PingOne Verify is as follows:

Capture document: The user captures images of the front and back of their identity document, typically using a mobile device.
Classify document: The system classifies the document based on type (passport, driver license, and so on) and region of origin, using machine learning models trained on various document types.
Compare to model: The captured document is compared against the pre-trained model for that document type. The model checks for specific security features such as watermarks, barcodes, and holograms that must match what’s expected for an authentic document.
Check for tampering or forgery: The system checks for signs of forgery or tampering, such as photocopied or altered documents, misaligned text, and improper security features.
Extract data: Relevant data (for example, name, date of birth, document number) is extracted using Optical Character Recognition (OCR) or, when available, directly from encoded barcodes like the PDF417 found on many driver licenses.
Deliver result: The result is delivered as either a successful authentication or a failure, accompanied by specific failure reasons if applicable (for example, “Document appears to be a photocopy” or “Expired ID”).

Types of document authentication

PingOne Verify offers multiple types of document authentication methods:

Document authentication method	Definition
Automated	Machine learning models handle the entire authentication process without human intervention. This is the fastest, most scalable option and is ideal for most standard use cases.
Manual	In specific situations where additional review is needed, manual verification can be performed by a human agent who reviews the document based on its visual features.
Step-Up	In high-risk scenarios or when automated verification fails, users could be required to perform a step-up authentication. This can include re-verifying their identity through an additional document authentication check after the initial process.

Document authentication method

Definition

Automated

Machine learning models handle the entire authentication process without human intervention.

This is the fastest, most scalable option and is ideal for most standard use cases.

Manual

In specific situations where additional review is needed, manual verification can be performed by a human agent who reviews the document based on its visual features.

Step-Up

In high-risk scenarios or when automated verification fails, users could be required to perform a step-up authentication.

This can include re-verifying their identity through an additional document authentication check after the initial process.

What is a false positive? What is a false negative?

In machine learning and document authentication, the terms false positive and false negative are important for understanding the accuracy of the system:

Term	Definition
False positive	Occurs when a fraudulent or altered document is incorrectly flagged as valid by the system. For example, if someone submits a doctored ID and the system verifies it as genuine, that is a false positive. This represents a security risk as unauthorized individuals could gain access to sensitive systems.
False negative	Occurs when a valid, legitimate document is incorrectly flagged as fraudulent or invalid. For example, if a user’s authentic ID is not verified due to poor image quality or unusual document features, this is a false negative. This can cause frustration for legitimate users trying to verify their identity.

Term

Definition

False positive

Occurs when a fraudulent or altered document is incorrectly flagged as valid by the system.

For example, if someone submits a doctored ID and the system verifies it as genuine, that is a false positive.

This represents a security risk as unauthorized individuals could gain access to sensitive systems.

False negative

Occurs when a valid, legitimate document is incorrectly flagged as fraudulent or invalid.

For example, if a user’s authentic ID is not verified due to poor image quality or unusual document features, this is a false negative.

This can cause frustration for legitimate users trying to verify their identity.

PingOne Verify minimizes both false positives and false negatives through continuous improvement of its machine learning models, though some level of both is expected because of the complex nature of document verification across various regions and document types.