[1]王新武,陈春雨.基于 U 形多尺度注意力方法的真实图像去噪[J].计算机技术与发展,2024,34(04):48-54.[doi:10. 3969 / j. issn. 1673-629X. 2024. 04. 008]
　WANG Xin-wu,CHEN Chun-yu.Real-world Image Denoising Based on U-shaped Multi-scale Attention Method[J].,2024,34(04):48-54.[doi:10. 3969 / j. issn. 1673-629X. 2024. 04. 008]

点击复制

基于 U 形多尺度注意力方法的真实图像去噪()

分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

34

期数:

2024年04期

页码:

48-54

栏目:

媒体计算

出版日期:

2024-04-10

文章信息/Info

Title:

Real-world Image Denoising Based on U-shaped Multi-scale Attention Method

文章编号:

1673-629X(2024)04-0048-07

作者:

王新武; 陈春雨

哈尔滨工程大学 信息与通信工程学院,黑龙江 哈尔滨 150001

Author(s):

WANG Xin-wu; CHEN Chun-yu

School of Information and Communication Engineering,Harbin Engineering University,Harbin 150001,China

关键词:

图像去噪; 计算机视觉; 真实噪声; 多尺度特征; 长距离通道注意力

Keywords:

image denoising; computer vision; real noise; multi-scale features; long-range channel attention

分类号:

TP391

DOI:

10. 3969 / j. issn. 1673-629X. 2024. 04. 008

摘要:

针对真实世界图像去噪算法存在对上下文信息和全局信息利用不足导致的去噪效果不佳问题,提出一种 U 形金字塔注意力网络( UPCA) 。 U 形结构由多尺度特征模块与长距离通

道注意力模块融合形成的金字塔注意力模块组成,U形结构通过拼接操作可以将每一层的输出特征图融合,减少卷积过程以及下采样过程中图像细节特征的丢失。 多尺度特征金字

塔模块可以更好地利用上下文信息从而更好地恢复出干净的图像,而建立长距离依赖的通道注意力模块可以更好地利用全局信息,提高网络的去噪效果。 同时在损失函数部分加入

噪声项来加快训练时收敛的速度以及提高去噪效果。UPCA 网络在数据集 SIDD 和 DND 进行对比实验,验证了 UPCA 网络的可行性和先进性,同时与同样使用通道注意力的RIDNet 相比 UPCA 网络的 PSNR / SSIM 指标提升了 0. 81 dB / 0. 044,去噪后的效果图直观表现也更好,而且同等参数下训练所需的算力更小。

Abstract:

To address the issue of subpar denoising results in existing algorithms for real - world image denoising, we propose aninnovative solution called the U-Shape Pyramid Channel Attention ( UPCA) . The U-shape structure comprises a fusion of multi-scalefeature modules and long-range channel attention modules,forming a pyramid attention module. Through concatenation operations,the U-shape structure allows for the fusion of output feature maps from each layer,minimizing the loss of fine-grained image details during theconvolution and downsampling processes. The multi-scale feature pyramid module effectively leverages contextual information to restoreclean images,while the long - range channel attention module establishes dependencies on global information, thereby enhancing thedenoising performance of the network. Additionally, we introduce a noise term in the loss function to expedite convergence duringtraining and improve denoising efficiency. Experimental comparisons on the SIDD and DND datasets demonstrate the feasibility and superiority of the UPCA. Compared to RIDNet which also utilizes channel attention,UPCA achieves a remarkable improvement of 0. 81dB / 0. 044 in terms of PSNR / SSIM metrics. The visually enhanced denoised images produced by UPCA are superior,and it requires lesscomputational power for training with the same set of parameters.

常用功能