Skip to content
arXiv· arXiv:2607.26976· Jian Xiong, Wenbo Jiang, Zihan Wang, Rui Zhang, Wenshu Fan, Hongwei Li, Guowen Xu·Indexed by this site · Originally published

InkShield: Handwriting Style Protection Against Unauthorized Handwriting Imitation

InkShield: Writing Style Protection Against Unauthorized Handwriting Mimicry

AI summary

InkShield protects handwriting reference images before release against unauthorized handwriting imitation. It selects decoy writers to define the direction of style shift, optimizes perturbations using a frozen handwriting generation surrogate model, and restricts perturbations to stroke edges; on IAM, two independent writer evaluators reduce the average Top-1/Top-5 retrieval of generated samples as the target writer from 11.94%/36.52% to 2.03%/8.79%, the LPIPS of protected images is 0.0078, and the generated text remains readable. The method can also transfer to other handwriting generators.

Read originalarxiv.org