Scholarlink

<p>Image smoothing plays a critical role in various image processing tasks such as detail enhancement, HDR tone mapping and non-photorealistic rendering. The global optimization model is an essential instrument for image smoothing, due to the excellent smoothing quality. However, most of them are based on static guidance, thus may struggle to balance between detail smoothing and edge preservation. In this paper, we propose a novel optimization model with dynamic guidance for image smoothing, which significantly improves the smoothing quality. To solve the proposed model, we proposed a novel iterative solution based on the plug and play framework, where the main computational burden in each iteration is a weighted least square model that can be efficiently solved with unsupervised learning. We have conducted extensive experiments to evaluate the proposed filter. Both quantitative and qualitative results demonstrate the superiority of the proposed method on various applications. Our filter is efficient, it is able to process 720P color images at interactive rates.</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

Deep unsupervised image smoothing with dynamic guidance

  • 期刊论文

  • 发表日期：

    2026年5月5日

查看全文
  • Yang Yang,
  • Yue Liu,
  • Xinyu Wang,
  • Xinsheng Wang

摘要

Image smoothing plays a critical role in various image processing tasks such as detail enhancement, HDR tone mapping and non-photorealistic rendering. The global optimization model is an essential instrument for image smoothing, due to the excellent smoothing quality. However, most of them are based on static guidance, thus may struggle to balance between detail smoothing and edge preservation. In this paper, we propose a novel optimization model with dynamic guidance for image smoothing, which significantly improves the smoothing quality. To solve the proposed model, we proposed a novel iterative solution based on the plug and play framework, where the main computational burden in each iteration is a weighted least square model that can be efficiently solved with unsupervised learning. We have conducted extensive experiments to evaluate the proposed filter. Both quantitative and qualitative results demonstrate the superiority of the proposed method on various applications. Our filter is efficient, it is able to process 720P color images at interactive rates.