Title: Extracting and Analyzing Handwritten Signatures in Swiss Popular Initiatives

Abstract: The validation of handwritten signature lists is a resource-intensive manual procedure, despite popular initiatives and referendums being pivotal components of Swiss democracy. This study explores the viability of automated document analysis techniques, such as Optical Character Recognition (OCR) and artificial intelligence-driven handwriting analysis, to assist in this process. We introduce a workflow that integrates template-based line segmentation with text recognition and writer retrieval methods, testing these approaches on a corpus comprising 443 handwritten samples from 418 distinct writers. Our findings reveal that OCR systems face significant challenges with out-of-vocabulary handwriting, achieving a Character Error Rate (CER) of 29.6% for first names. Conversely, writer retrieval demonstrated greater resilience, attaining a mean Average Precision (mAP) of 50.6%. The experiments suggest that commercial, off-the-shelf OCR solutions lack the necessary reliability for transcribing handwritten signature data, especially concerning brief, non-standard entries like names and addresses. Nevertheless, writer retrieval techniques prove effective at identifying visually analogous entries across different signature lists, positioning them as a valuable asset for detecting potential duplicate submissions through handwriting similarity.

Source: arXiv Generated at: 2026-06-04 00:00:00 UTC