A Machine Learning Framework for Forgery Detection in Digital ID Documents

This paper introduces an innovative document verification system that leverages artificial intelligence techniques to simplify the onboarding process of electronic identity solutions. The system uses images of identity documents captured via smartphones to identify and detect any potential manipulation or alterations present within the documents. It addresses the wide variety of document types and versions, and the variability in image quality due to different smartphone cameras and lighting conditions. The technological stack used for identity document verification includes optical character recognition (OCR) libraries, machine-readable zone (MRZ) check, image key-points for copy-move forgery detection, and advanced machine learning algorithms for character manipulation detection. The paper also describes the dataset used for training and validation, consisting of genuine identity documents and simulated forged documents. The verification module includes an image quality check, a copy-move forgery detection, an imitation forgery detector, and a global forgery scoring endpoint. This system provides a comprehensive approach for real-time verification of valid identity documents, which has been tested across five European countries, offering a transparent and secure framework to detect forged identity documents.