Scholarlink

Self similarity modeling, derived from fractal analysis, is crucial in image super resolution. However, magnifying images significantly often introduces “Facet noise,” where local pixels show an irregular distribution like fish scales, compromising image aesthetics and obscuring details. To address this, we present a range of solutions, including Hermite matrix decomposition for structural analysis, directional filter banks for enhancing local texture, truncated singular value decomposition to reduce noise, and an energy loss function to guide model convergence. We also explore self similarity and matrix self similarity models to improve image quality, collectively termed as Facet. Rigorous experiments on renowned datasets show that Facet effectively suppresses Facet noise, restores natural textures, and enhances the image quantization index, demonstrating its practicality and contribution to image super resolution research.

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

Facet: Bag of Tricks to Suppress Facet Noise in Image Super Resolution Based on Self Similarity

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • He Jiang,
  • Mang Sun,
  • Chentao Zhu,
  • Xin Huang,
  • Long Li,
  • Yaoman Huang,
  • Jieheng Li,
  • Peilin Wu,
  • Yushan Chen,
  • Hanzheng Guo,
  • Boming Song,
  • Deqiang Cheng

摘要

Self similarity modeling, derived from fractal analysis, is crucial in image super resolution. However, magnifying images significantly often introduces “Facet noise,” where local pixels show an irregular distribution like fish scales, compromising image aesthetics and obscuring details. To address this, we present a range of solutions, including Hermite matrix decomposition for structural analysis, directional filter banks for enhancing local texture, truncated singular value decomposition to reduce noise, and an energy loss function to guide model convergence. We also explore self similarity and matrix self similarity models to improve image quality, collectively termed as Facet. Rigorous experiments on renowned datasets show that Facet effectively suppresses Facet noise, restores natural textures, and enhances the image quantization index, demonstrating its practicality and contribution to image super resolution research.