摘要
近年来,在深度卷积神经网络对真实图像去噪的研究中,现有方法可以一定程度地去除真实图像噪声,但由于真实噪声的随机性和复杂性,使得模型在处理较为复杂的真实噪声时仍有一定的局限性,在还原图像细节与关键特征信息的提取方面表现一般。针对以上问题,在CBDNet的基础上提出了一种基于注意力与特征融合的去噪算法。具体来说,在上采样过程中通过添加空间与通道融合的注意力机制来提取更多相关特征,并重新定义跳跃连接的输入进行特征融合。在下采样过程中使用最大池化替代平均池化来增强图像纹理和细节。实验方面,在SIDD、NC12、Nam等三个真实噪声数据集上测试并与多个先进算法进行对比,实验结果表明了该算法在定量和视觉上的优越性。
This paper proposes a denoising algorithm based on attention and feature fusion based on CBDNet.Specifically,in the upsampling process,more relevant features are extracted by adding the attention mechanism of space and channel fusion,and the input of jump connection is redefined for feature fusion.Use maximum pooling instead of average pooling during downsampling to enhance image texture and detail.In terms of experiment,this paper tests on three real noise data sets,such as SIDD,NC12 and Nam,and compared them with several advanced algorithms.The experimental results show that the algorithm is superior in quantitative and visual aspects.
出处
《工业控制计算机》
2024年第3期59-61,共3页
Industrial Control Computer
关键词
真实图像去噪
注意力机制
跳跃连接
池化
real image denoising
attention mechanism
skip connection
pooling