A Hybrid and Explainable Framework for Document Image Forgery Detection and Offline Signature Verification

Digital documents are extensively used in law, academics, and finance, and that is why they are considered extremely susceptible to image manipulations and offline handwritten signature forging. Most of the existing approaches view document image forgery detection and signature verification as independent problems, and most of the approaches provide a binary response rather than the location and nature of the alteration. The paper proposes an explanation and hybrid authentication, thus document authentication, through the integration of both cryptographic integrity check and conventional computer vision-based forensic evidence. An integrity check, MD5, is a hash-based check that is first applied to files to quickly detect changes and avoid unnecessary processing of authentic files. To localize forged areas in case of mismatch, OpenCV-based procedures are applied, which are based on copy-move detection, splicing analysis, and noise and texture inconsistency. Isolating and performing offline signature verification are also done in the model by using the assistance of the geometric, texture, and gradient-based features. The given solution provides a simple forensic evidence based on the visual explainability and lower cognition complexity, and can be applied to the real-life document verification models.