Glossary

Document fraud comes with its own jargon, and so does BlinkID Verify. Depending on your business, these terms might be more or less familiar, but it's always good to have a quick reference.

Fraud

In the context of BlinkID Verify, by default, fraud is the act of presenting a document that is not genuine or live. Verify's job is to detect this, but it also provides flexibility for you to define fraud for your company's or client's use cases. For example, you can disable liveness checks, and only check for the document being genuine.

(Document) Liveness

A "live" document is one that's physically present with the person presenting it, in its original form (not printed, photocopied, etc). The term liveness is used both as a property of documents, and for the act of checking for this property.

Verification

A verification is a single transaction that determines whether a document is genuine and live. It can be performed on one or multiple images depending on the document, capture conditions, and your desired options.

Verifiable

A verifiable image is one that satisfies our image requirements, contains a supported document, and can be verified by BlinkID Verify in such a way as to result in a Accept, Reject, or ManuallyReview RecommendedOutcome.

Options/Settings

Options (previously referred to as settings) are the various individual configurations for a single verification. They can be related to the behavior of the actual fraud detecting engine, capturing process, or operational things like whether the API should return images or not.

Check

A check is an abstract representation of single semantic unit of fraud detection. It makes sense by itself, but can also contain other checks. A single "check" in the result doesn't mean that a single operation was performed inside of BlinkID Verify and some checks represent hundreds of validations, or multiple ML models.

CertaintyLevel

CertaintyLevel is the measure of confidence that BlinkID Verify has in some of its checks. It can be Low, Medium, or High (or NotPerformed if the check was not performed). Some checks are inherently less certain (you can never be 100% sure someone is not actually called Sample Specimen) so when they fail, they never fail with a High certainty level.

Some checks can rely on multiple factors, and the presence of more of them increases the certainty level, while the presence of too few can decrease it.

A lower certainty level is usually a sign manual review might be needed.

MatchLevel

MatchLevels are our mechanism for configuring and giving rich results from checks that are on a gradient. Most of the checks that use this mechanism are ML models.

You can use a MatchLevel to configure the sensitivity of a check. For example, MatchLevel 8 is more strict than MatchLevel 4. A MatchLevel can also be set to Disabled to completely disable the check.

On the response side, all checks configurable with MatchLevels will also return a satisfied MatchLevel that indicates how strongly the check passed or failed.

(Process)Indicator

Process indicators are pieces of information about a single verification that don't pertain to fraud per se, but can affect, influence or even completely block the fraud verdict, and are included in the result to provide explanations and context. This includes things like image quality issues, improper cropping, etc.

Manual Review

Manual review is the act of a human expert or worker reviewing a document. This review can be detailed, and check for fraud, but it can also be a simple check due to inconsistencies, or minor issues.

BlinkID Verify is a fully automated solution, but it can sometimes recommend manual review, if your process allows for it. The sensitivity of this recommendation is configurable, and can be adjusted to your needs.

Front and back images

Most identity documents are double sided, and have a front and back side. For BlinkID Verify, "front image" typically refers to a full frame with context that contains a document's front side. "Back image" typically refers to a full frame with context that contains a document's back side.

For single sided documents, or documents like passports, we usually refer to the relevant image as the "front image".

Extraction

Extraction is the process of extracting personal data from a document. We also sometimes use it as a noun, and "an extraction" is a single transaction performed on one or more images of one document that extracts the data from it from various sources (VIZ, MRZ, barcodes, images, etc).

Every verification requires at least some degree of extraction.

FAR

False acceptance rate, or FAR, is the rate at which a verification solution incorrectly accepts fraudulent documents as legitimate.

FAR is sometimes also known as APCER (Attack Presentation Classification Error Rate).

FRR

False rejection rate, or FRR, is the rate at which a verification solution incorrectly rejects legitimate documents as fraudulent.

FRR is sometimes also known as BPCER (Bona Fide Presentation Classification Error Rate)

TAR

True acceptance rate, or TAR, is the rate at which a verification solution correctly accepts legitimate documents as legitimate. For a solution that only has a boolean output, this is equivalent to 1-FRR. However, if the solution has outputs other than PASS/FAIL (which BlinkID Verify does), it's its own metric.

TRR

True rejection rate, or TRR, is the rate at which a verification solution correctly rejects fraudulent documents as fraudulent. For a solution that only has a boolean output, this is equivalent to 1-FAR. However, if the solution has outputs other than PASS/FAIL (which BlinkID Verify does), it's its own metric.