基于联结扩张注意网络的图像去摩尔纹算法
Image Demoireing Algorithm Based on Associative Extended Attention Network
孙光灵 1卢慧敏 2陈冲 2黄磊 2苏亮亮2
作者信息
- 1. 安徽建筑大学 电子与信息工程学院,安徽 合肥 230601;合肥工业大学 智能互联系统安徽省实验室,安徽 合肥 230002
- 2. 安徽建筑大学 电子与信息工程学院,安徽 合肥 230601
- 折叠
摘要
针对相机拍摄数字显示屏时图像产生的摩尔纹,本文提出了基于联结扩张注意网络的图像去摩尔纹算法.该算法先以轻量级基线U-Net网络为基础,将改进的动态特征金字塔模块DFP引入U-Net的跳跃连接中,并构成联结结构,实现多尺度特征提取与融合,再利用CBAM改进的软注意力机制RMAM来聚焦摩尔纹图像颜色信息,最后构建网络模型对其进行了训练及测试.结果表明,相较于其他算法,该算法具有较高的峰值信噪比和结构相似度.
Abstract
In order to solve the problem of moire patterns in digital display images taken by cameras, a demoireing algo-rithm based on connected extended attention network is proposed. Firstly, based on lightweight baseline U-Net network, dy-namic feature pyramid (DFP) module is introduced into U-Net skip connections to construct the connection structure and real-ize multi-scale feature extraction and fusion. Secondly, an improved soft attention mechanism based on CBAM, RMAM is used to focus the color information of moire images. Finally, the network model is built on the above foundation, and it is trained and tested. Experimental results show that the proposed algorithm achieves a higher peak signal-to-noise ratio and structural similarity compared to other existing algorithms.
关键词
图像处理/图像去摩尔纹/U-Net网络/多尺度/软注意力机制Key words
image processing/image demoireing/U-Net network/multi-scale/soft attention mechanism引用本文复制引用
基金项目
国家自然科学基金项目(62001004)
安徽省高校协同创新项目(GXXT-2021-024)
安徽省住房城乡建设科学技术计划项目(2023-YF058)
安徽省住房城乡建设科学技术计划项目(2023-YF113)
合肥工业大学"智能互联系统安徽省实验室"开放基金(PA2021AKSK0107)
新时代育人质量项目(2023zyxwjxalk112)
出版年
2024