基于边缘引导与特征融合的分阶段图像去模糊
CSTR:
作者:
作者单位:

(西安建筑科技大学 理学院,陕西 西安 710055)

作者简介:

陈清江 (1966-),男,博士,教授,硕士生导师,主要从事小波分析、图像处理与信号处理方面的研究。

通讯作者:

中图分类号:

TP391.41

基金项目:

国家自然科学基金(12202332,61902304)和陕西省自然科学基础研究计划(2021JQ-495) 资助项目


Phased image deblurring based on edge guidance and feature fusion
Author:
Affiliation:

(College of Science, Xi′an University of Architecture and Technology, Xi′an, Shaanxi 710055, China)

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    针对现有方法去模糊后图像缺乏足够清晰边缘的问题,提出了一种基于边缘引导与特征融合的分阶段图像去模糊方法,分两个阶段逐步去除模糊。首先,第一阶段应用含双交叉集成注意力模块(dual cross integrated attention module,DCIAM)的编解码网络学习不同尺度下图像的内容特征,初步去除模糊信息。然后,构建边缘分支网络(edge branch network,EBN)提取图像边缘特征。再次,设计了边缘引导去模糊模块(edge-guided deblurring module,EGDM)以耦合不同分辨率下图像的内容与边缘特征。最后,第二阶段通过级联的残差块和DCIAM进一步去除图像模糊,并通过引入自校准注意融合模块(self-calibrating attention fusion module,SCAFM)增强特征表达效果。实验结果表明:所提方法去模糊后图像的峰值信噪比和结构相似度均值分别达到32.78 dB和 0.964 ,均优于其他对比方法。所提方法可以显著提高去模糊性能且去模糊后图像边缘结构更加完整。

    Abstract:

    Aiming at the problem that the image lacks sufficient clear edge after deblurring by existing methods,a phased image deblurring method based on edge guidance and feature fusion is proposed,and the deblurring task is divided into two stages to gradually remove blur.Firstly,the codec network with double cross integrated attention module (DCIAM) is used to learn the content features of images at different scales to realize the preliminary removal of blur.Secondly,an edge branch network (EBM) is constructed to extract image edge features.Thirdly,an edge-guided deblurring module (EGDM) is designed to couple the content and edge features of images at different resolutions.Finally,cascaded residual blocks and DCIAMs are used to achieve further remove of blur,and a self-calibrated attention fusion module (SCAFM) is introduced to enhance the feature expression.The experimental results demonstrate that the average peak signal-to-noise ratio and structural similarity of the proposed method reach 32.78 dB and 0.964,respectively,which are superior to other comparison methods.The proposed method can significantly improve the deblurring performance and make the image edge structure more complete after deblurring.

    参考文献
    相似文献
    引证文献
引用本文

陈清江,邵菲,王炫钧.基于边缘引导与特征融合的分阶段图像去模糊[J].光电子激光,2025,(1):17~26

复制
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2023-05-13
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2024-12-11
  • 出版日期:
文章二维码