Detects visual artifacts like logos, seals, graphs, images, or stamps within document layouts.
Finds and classifies the non-text content that carries meaning: signatures, stamps, seals, logos, handwriting, barcodes. Often the part of a document that establishes its authenticity rather than its content.
Presence or absence of a visual element is itself the fact needed — is this signed, is it stamped, does it carry the right seal.
Where the visual element is decorative. Detecting a letterhead logo proves the paper, not the claim.
Detection must not be conflated with verification. Finding a signature-shaped mark is not evidence that the right person signed, and systems that blur the two create false assurance.
High false-confidence on tampered or low-quality scans. Forgery detection in particular tends to produce either many false positives or a comfortable silence, and tuning between the two is genuinely hard.
An insurer checking that a claim form carries a signature in the required field before it enters the assessment queue — a completeness check, not an identity check.