Scholarlink

<p>Neural Radiance Fields (NeRF) excel in novel view synthesis but face challenges when handling scenes with reflective materials due to complex viewpoint-dependent reflections. Existing methods often rely on manual segmentation or overly simplified reflection models, limiting their effectiveness. To address this, we propose RAS-NeRF, a framework that integrates geometric constraints and Fourier feature mapping to dynamically adjust frequency information, enhancing reflection perception. This approach effectively removes interference from reflective surfaces such as glass covers, achieving high-quality reflection-free image synthesis. Our method not only improves novel view rendering quality but also enhances depth estimation accuracy, thereby strengthening the model’s scene understanding in complex environments. When applied to digital cultural heritage protection, it can effectively remove reflection artifacts and ensure the details of cultural heritage. This provides strong technical support for 3D reconstruction and display, making the protection and display of cultural heritage more accurate and reliable.</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

RAS-NeRF for novel view synthesis of complex reflective artifacts

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年5月12日

查看全文
  • Yizhi Wang,
  • Dong Sui,
  • Yacong Li,
  • Yang Li,
  • Maozu Guo

摘要

Neural Radiance Fields (NeRF) excel in novel view synthesis but face challenges when handling scenes with reflective materials due to complex viewpoint-dependent reflections. Existing methods often rely on manual segmentation or overly simplified reflection models, limiting their effectiveness. To address this, we propose RAS-NeRF, a framework that integrates geometric constraints and Fourier feature mapping to dynamically adjust frequency information, enhancing reflection perception. This approach effectively removes interference from reflective surfaces such as glass covers, achieving high-quality reflection-free image synthesis. Our method not only improves novel view rendering quality but also enhances depth estimation accuracy, thereby strengthening the model’s scene understanding in complex environments. When applied to digital cultural heritage protection, it can effectively remove reflection artifacts and ensure the details of cultural heritage. This provides strong technical support for 3D reconstruction and display, making the protection and display of cultural heritage more accurate and reliable.