一种基于汉字字形扰动的字体风格迁移方法
By constructing a small dataset of Chinese character glyph perturbations and a style transfer network, and combining attention-enhanced convolutions and transposed convolutional blocks, the problem of low efficiency in Chinese character generation is solved, and efficient generation of Chinese character variants and covert communication applications are realized.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HANGZHOU DIANZI UNIV
- Filing Date
- 2022-11-22
- Publication Date
- 2026-07-17
AI Technical Summary
The large size and diverse font styles of Chinese character datasets lead to low efficiency in generating glyph perturbations and Chinese character variants, and existing technologies lack effective methods for font style transfer.
By constructing a small dataset of Chinese character glyph perturbations, a style transfer network is used to extract the skeleton and style features of the dataset. Attention-enhanced convolutions and transposed convolutional blocks are combined to generate variant images of the target font Chinese characters. The generation efficiency is improved by training an adversarial discriminator.
The generated target font Chinese character variants have clear strokes and complete structures, with high recognizability and authenticity, making them suitable for covert communication.
Smart Images
Figure CN115688677B_ABST