Validation study demonstrates reliable forged document detection in mobile-captured identity records, indicating lightweight explainable models can secure identity verification pipelines.
Key Points
To develop and validate an explainable, lightweight AI framework for authenticating identity documents against visual and semantic manipulations in AML and KYC compliance environments.
Integrated handcrafted forensic feature extraction, optical character recognition (OCR)-based semantic field extraction, and Random Forest classification.
Trained and evaluated the system on selected MIDV-2020 subsets containing Albanian identity cards, Latvian passports, and Slovakian identity cards photographed under realistic mobile imaging conditions.
Achieved a recall rate of 92.31% on the held-out test set.
Attained an overall authentication accuracy of 84.85% while providing interpretable feature outputs without requiring computationally heavy deep learning models.