Handwritten Chinese Signature Detection on Scanned Technical Documents for Authenticity Verification

Signature detection is one of the crucial predecessor procedures to prove the authenticity of handwritten signatures. Despite the increasing demand for automatedly verifying the signatures on contracts, bills, reports, etc., the research based on signature detection is very limited. One of the impediments to signature detection is the lack of public annotated datasets due to the interests of privacy and confidentiality. In this paper, we explore a data augmentation method, the Copy-Paste augmentation, to alleviate the scarcity of signed documents. Modeling signature detection as an object detection task, we experiment on Chinese technical documents using different object detection models. The experiment shows that YOLOv5 with Copy-Paste augmentation performs best. Based on the experiment, we propose a prototype system for signature detection.