Signature Detection, Restoration, and Verification: A Novel Chinese Document Signature Forgery Detection Benchmark

Offline signature forgery detection has attracted many researchers in recent years. In real situations, signatures should be detected from the signed documents and verified by the forgery detection system. There are many challenges in the pipeline. First, some signatures have low resolutions and are difficult to be detected. Second, the cropped signatures may contain irrelevant background context of the document, making the signature hard to be verified. Third, some forgery signatures are very similar to genuine ones, increasing the challenge of verification. In addition, most existing datasets do not cover all the pipeline tasks. Moreover, publicly available Chinese-based signature datasets are rare for research purposes. In this paper, we construct a novel Chinese document offline signature forgery detection benchmark, namely ChiSig, which includes all pipeline tasks, i.e., signature detection, restoration, and verification. Besides, we extensively compare different deep learning-based approaches in these three tasks. The results show that our proposed dataset can effectively provide solutions for constructing pipeline systems for Chinese document signature forgery detection.