SDGDiff: A spectral-domain guided diffusion model for underwater image super-resolution

被引：0

作者：

Xin Liu ^{[1
]}

Kaijie Zhang ^{[2
]}

Zhiqiu Yan ^{[3
]}

Sipei Huang ^{[1
]}

Zhengxiang Gu ^{[1
]}

Yi Zhang ^{[3
]}

Li Wang ^{[4
]}

机构：

[1] Jinken College of Technology,

[2] Suzhou University,undefined

[3] Southeast University,undefined

[4] Nanjing Vocational University of Industry Technology,undefined

来源：

Signal, Image and Video Processing | 2025年 / 19卷 / 9期

关键词：

Diffusion Model; Super-Resolution; Spectral-Domain;

D O I：

10.1007/s11760-025-04373-4

中图分类号：

学科分类号：

摘要：

Improving the quality of underwater images is crucial for ocean resource exploration. However, due to the influence of light attenuation and scattering, although existing methods can achieve high evaluation metrics scores, there are still shortcomings in detail restoration and color reproduction, making it difficult to improve visual perception. Therefore, this article proposes a spectral domain guided diffusion model SDGDiff, which enhances visual perception while maintaining high scores. This method adopts a two-stage strategy for underwater image super-resolution reconstruction: the first stage is the coarse restoration stage, responsible for image enlargement and color correction; The second stage is the spectral domain guidance stage, which utilizes a diffusion model to compensate for high-frequency details in the residual space, thereby enhancing the expressive power of details. In addition, by combining the loss function based on discrete wavelet transform, the ability to recover details is further optimized. The experimental results indicate that SDGDiff achieves a good balance between image quality and visual perception.

引用