Journal of Beijing Normal University, Volume. 61, Issue 3, 285(2025)

Uncertainty-awared generative network for Chinese scene text editing

GAO Yutong1,2, ZHANG Ying1, LIU Xianggan3,4、*, LIU Yidian5, JIANG Shan5, GUO Ziyi5, and SONG Feifan6
Author Affiliations
  • 1Key Laboraory of Ethnic Language Intelligent Analysis and Security Governance, Ministry of Education, Minzu University of China, Beijing, China
  • 2Key Laboratory of Big Data and Artificial Intelligence in Transportation, Ministry of Education, Beijing Jiaotong University, Beijing, China
  • 3Natural Language Processing and Knowledge Graph Laboratory, Huazhong University of Science and Technology, Wuhan, Hubei, China
  • 4Hainan Lingshui Li'an International Education Innovation Pilot Zone, Lingshui, Hainan, China
  • 5State Key Laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications, Beijing, China
  • 6Research Center for Rural Financial Reform, Changchun, Jilin, China
  • show less

    A Chinese scene text editing (CSTE) method, based on research but incorporating uncertainty modeling and identifying effective technical solution, is proposed in this work. This new method optimizes prediction noise through an uncertainty-guided adjustment mechanism, improving the accuracy of noise estimation, thereby enhancing the clarity and structural integrity of the generated text. Additionally, by filtering irrelevant information from both textual and visual features, the method improves cross-modal alignment capabilities, achieving a seamless fusion of text and background textures.

    Tools

    Get Citation

    Copy Citation Text

    GAO Yutong, ZHANG Ying, LIU Xianggan, LIU Yidian, JIANG Shan, GUO Ziyi, SONG Feifan. Uncertainty-awared generative network for Chinese scene text editing[J]. Journal of Beijing Normal University, 2025, 61(3): 285

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Received: Apr. 9, 2025

    Accepted: Aug. 21, 2025

    Published Online: Aug. 21, 2025

    The Author Email: LIU Xianggan (liuxg@hust.edu.cn)

    DOI:10.12202/j.0476-0301.2025056

    Topics