🤖 AI Summary
Offline handwritten signature verification faces core challenges including high inter-writer similarity, large intra-writer variability, and severe scarcity of labeled samples. To address these, we propose Multi-Scale Signature Network (MS-SigNet) coupled with a novel Co-Tuplet Loss—marking the first approach to jointly model global structural layout and local fine-grained details in signature verification. Our method integrates multi-scale feature extraction, metric learning, and end-to-end optimization to enhance discriminative representation learning via contrastive learning over multiple positive and negative samples. Additionally, we construct and publicly release HanSig, the first large-scale Chinese signature dataset. Extensive experiments demonstrate that our method significantly outperforms state-of-the-art methods across four cross-lingual benchmarks, particularly achieving substantial accuracy gains under highly similar forgeries, and exhibits strong generalization capability.
📝 Abstract
Handwritten signature verification, crucial for legal and financial institutions, faces challenges including inter-writer similarity, intra-writer variations, and limited signature samples. To address these, we introduce the MultiScale Signature feature learning Network (MS-SigNet) with the co-tuplet loss, a novel metric learning loss designed for offline handwritten signature verification. MS-SigNet learns both global and regional signature features from multiple spatial scales, enhancing feature discrimination. This approach effectively distinguishes genuine signatures from skilled forgeries by capturing overall strokes and detailed local differences. The co-tuplet loss, focusing on multiple positive and negative examples, overcomes the limitations of typical metric learning losses by addressing inter-writer similarity and intra-writer variations and emphasizing informative examples. We also present HanSig, a large-scale Chinese signature dataset to support robust system development for this language. The dataset is accessible at url{https://github.com/hsinmin/HanSig}. Experimental results on four benchmark datasets in different languages demonstrate the promising performance of our method in comparison to state-of-the-art approaches.