Scholarlink

<p>Laparoscopy is crucial for diagnosing and surgically treating intra-abdominal diseases. However, due to the hardware limitations of laparoscopic devices, the acquired images often suffer from low resolution, blurred details, and unclear edge structures, which hinder precise assessment of pathological features. To address this issue, this paper proposes a Gradient Collaborative Guidance Super-Resolution (GCGSR) method for structure-consistent laparoscopic image reconstruction. The proposed network adopts a dual-pathway architecture consisting of a super-resolution reconstruction branch and a gradient-guided branch. Specifically, the SR branch performs progressive feature learning and hierarchical upsampling through multiple feature extraction modules to achieve the initial reconstruction of image content. Meanwhile, the gradient branch takes the gradient map of the low-resolution image as input and employs Gradient-Enhanced Convolution (GEConv) together with Gradient-Guided Attention (GGA) in the Gradient-Enhanced Attention Block (GEAB) to strengthen feature learning and generate high-resolution gradient maps. Structural priors obtained from both branches are then fused through bidirectional collaborative interaction, enhancing the network’s ability to learn edges, textures, and other critical structures. In addition, a Self-Supervised Structural Collaborative (SSSC) loss is introduced to constrain edge information and prevent information decoupling between the two branches. Experimental results demonstrate that the proposed method outperforms existing mainstream SR approaches on self-built Laparoscopic datasets in terms of PSNR, LPIPS, FID, and PI, while exhibiting superior structural fidelity.</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

Structure-consistency super-resolution based on gradient collaborative guidance for laparoscopic images

  • 期刊论文

  • 发表日期：

    2026年7月6日

查看全文
  • XiaoYing Pan,
  • Zicheng Lu,
  • Yan Wang,
  • Xingzhao Pan,
  • Tianxin Zhang

摘要

Laparoscopy is crucial for diagnosing and surgically treating intra-abdominal diseases. However, due to the hardware limitations of laparoscopic devices, the acquired images often suffer from low resolution, blurred details, and unclear edge structures, which hinder precise assessment of pathological features. To address this issue, this paper proposes a Gradient Collaborative Guidance Super-Resolution (GCGSR) method for structure-consistent laparoscopic image reconstruction. The proposed network adopts a dual-pathway architecture consisting of a super-resolution reconstruction branch and a gradient-guided branch. Specifically, the SR branch performs progressive feature learning and hierarchical upsampling through multiple feature extraction modules to achieve the initial reconstruction of image content. Meanwhile, the gradient branch takes the gradient map of the low-resolution image as input and employs Gradient-Enhanced Convolution (GEConv) together with Gradient-Guided Attention (GGA) in the Gradient-Enhanced Attention Block (GEAB) to strengthen feature learning and generate high-resolution gradient maps. Structural priors obtained from both branches are then fused through bidirectional collaborative interaction, enhancing the network’s ability to learn edges, textures, and other critical structures. In addition, a Self-Supervised Structural Collaborative (SSSC) loss is introduced to constrain edge information and prevent information decoupling between the two branches. Experimental results demonstrate that the proposed method outperforms existing mainstream SR approaches on self-built Laparoscopic datasets in terms of PSNR, LPIPS, FID, and PI, while exhibiting superior structural fidelity.