Elegantly Written V2: Next-Scale Prediction for Enhancing Online Chinese Handwriting

Yu Liu, Yuqiu Kong, Yang Ding, Fang Liu, Lei Wang, Cunrui Wang

Abstract: Electronic writing tools and environments, while enhancing writing efficiency and convenience, also compromise the readability and personalized style of handwritten characters due to the absence of mechanical feedback between the pen tip and the paper. To address this problem, this paper proposes a handwritten trajectory reconstruction method based on content-style decoupling and reorganization. Inspired by the human “global-first, local-second” hierarchical writing and perception process, we model the reconstruction task as a multi-scale, progressively refined procedure. The process is divided into two stages: In the content preservation stage, original handwritten parameters are used as content guidance to maintain the consistency of character content. In the style aggregation stage, reconstructed handwritten trajectories at different scales are integrated with both sequence and image modality styles. Qualitative and quantitative experiments on the CASIA-OLHWDB dataset, simulating imperfect handwriting, demonstrate the superiority of the proposed method.

External IDs:dblp:conf/prcv/LiuKDLWW25