Scholarlink

<p>Recently, the utilization of Quick Response (QR) codes has expanded significantly, especially in secure mobile payment platforms and interactive digital advertising. The compositions of black and white blocks are not only monotonous and lacking in aesthetics, but also lack personalized patterns which can be easily counterfeited. The existing algorithms enhanced QR codes by incorporating style transfer and diffusion generation-based approaches. However the recognition is improved significantly through generating beautiful colored QR codes, the decoding rate is often reduced. Furthermore, the method of beautifying QR codes does not consider anti-counterfeiting issues and requires manual identification of whether QR codes have been replaced. To address these issues, this paper introduces an end-to-end diffusion multi-style based QR code beautification framework that combines digital watermark embedding and authentication, ensuring dual protection for QR code targeted information. More specifically, ControlNet is pre-trained with a self-constructed dataset and also used to guide the text-driven diffusion model to generate aesthetically pleasing QR codes with preserved structural integrity. A differentiable sampling simulation layer is employed to enhance the decoding rate while achieving visually appealing results. Furthermore, an improved HiDDeN network embeds a fixed invisible watermark to verify the authenticity of the beautified QR code. Extensive experiments demonstrate that the proposed framework achieves a balance between watermark quality, QR code decoding rate, and image quality in various practical applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

BeautyMark: A diffusion model for aesthetic QR code generation with robust watermark authentication

  • 期刊论文

  • 发表日期：

    2026年3月30日

查看全文
  • Jiayi Xu,
  • Penghui Li,
  • Hangpeng Ren,
  • Jianfeng Lu,
  • Li Li,
  • Mahmoud Emam

摘要

Recently, the utilization of Quick Response (QR) codes has expanded significantly, especially in secure mobile payment platforms and interactive digital advertising. The compositions of black and white blocks are not only monotonous and lacking in aesthetics, but also lack personalized patterns which can be easily counterfeited. The existing algorithms enhanced QR codes by incorporating style transfer and diffusion generation-based approaches. However the recognition is improved significantly through generating beautiful colored QR codes, the decoding rate is often reduced. Furthermore, the method of beautifying QR codes does not consider anti-counterfeiting issues and requires manual identification of whether QR codes have been replaced. To address these issues, this paper introduces an end-to-end diffusion multi-style based QR code beautification framework that combines digital watermark embedding and authentication, ensuring dual protection for QR code targeted information. More specifically, ControlNet is pre-trained with a self-constructed dataset and also used to guide the text-driven diffusion model to generate aesthetically pleasing QR codes with preserved structural integrity. A differentiable sampling simulation layer is employed to enhance the decoding rate while achieving visually appealing results. Furthermore, an improved HiDDeN network embeds a fixed invisible watermark to verify the authenticity of the beautified QR code. Extensive experiments demonstrate that the proposed framework achieves a balance between watermark quality, QR code decoding rate, and image quality in various practical applications.