基于通道自注意力机制的图像降噪算法  

Image Denoising Algorithm via Channel-Wise Self-Attention Mechanism

作　　者：吴畏 吴昊[2] WU Wei;WU Hao(Department of Economic and Trade,Anhui Business and Technology College,Hefei 231131,China;Department of Computer and Artificial Intelligence,Hefei Normal University,Hefei 230601,China)

机构地区：[1]安徽工商职业学院经济贸易学院,安徽合肥231131 [2]合肥师范学院计算机与人工智能学院,安徽合肥230601

出　　处：《长春师范大学学报》2025年第2期42-50,共9页Journal of Changchun Normal University

基　　金：2022年度安徽省高校科研重大项目“全球价值链重构背景下我国中西部地区价值链功能升级路径研究”(2022AH040343);2023年度安徽高校自然科学研究项目“基于内容的视频信息检索计算模型研究”(2023AH051306);2020年度光电探测科学与技术安徽省高校联合重点实验室项目“基于虚拟线圈的雾天交通视频违规检测”(2020GDTC01)。

摘　　要：基于卷积神经网络的图像降噪算法对图像自身信息的利用率相对有限,导致降噪后的图像出现一定程度的失真。Transformer架构中的自注意力机制能够有效利用图片自身信息,但计算量相对较高,对内存空间依赖相对较大。针对卷积神经网络信息利用率相对有限以及Transformer计算量大的缺陷,提出一种基于通道自注意力机制的图像降噪算法,该算法能够有效利用图片自身信息,降低Transformer的计算量。实验结果表明,本文算法能够在保证降噪效率的基础上有效提升图片降噪效果,在与主流降噪算法的对比中也能取得较优的降噪效果。The image denoising algorithm based on convolutional neural network has relatively limited utilization rate of the image s own information,which leads to a certain degree of distortion of the denoised image.The self-attention mechanism in Transformer architecture can effectively wtilize the information of the image itself,but the calculation amount is relatively high and it demands substential memory space.To address such limitations of convolutional neural network and Transformer,an image denoising algorithm based on channel-wise self-attention mechanism is proposed,which can effectively utilize the information of the picture itself on the one hand and reduce the calculation amount of Transformer on the other.The experimental results show that the proposed algorithm can effectively improve the image denoising effect on the basis of ensuring the denoising efficiency.Compared with the mainstream noise reduction algorithms,it can also achieve better noise reduction effect.

关 键 词：图像降噪 卷积神经网络 TRANSFORMER 通道自注意力 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]