深度卷积神经网络(Convolutional Neural Network,CNN)在图像去噪领域受到广泛关注.然而,随着网络深度的增加,大多数深度CNN会出现性能饱和、学习能力下降等问题.提出了一种结合局部和全局特征的双路径去噪网络,将两个不同结构的网络组合后构成一个双路径模型,增加网络的宽度,从而获得更多不同的特征.通过长路径连接融合全局和局部特征,增强层间相关性.注意力机制利用当前阶段引导前一阶段的输入,获得更多的特征.实验结果表明,我们提出的网络模型在Set12和BSD68两个数据集中的峰值信噪比(Peak Signal-to-Noise Ratio,PSNR)分别达到 了 32.95 dB 和 31.74dB.同时,主观视觉效果(如图像的边缘等细节)恢复得更好、更清晰.
Dual-Path Image Denoising Network Based on Spatial Feature Fusion
Deep convolutional neural networks(CNNs)have attracted much attention in the field of image de-noising.However,with the increase in network depth,most deep CNNs have problems such as performance saturation and learning decline.In this paper,a dual-path denoising network combining local and global fea-tures is proposed.Two networks with different structures are combined to form a dual path model,and the width of the network is increased to obtain more different features.The global and local features are integrated through long path connections to enhance interlayer correlation.The attention mechanism uses the current stage to guide the input of the previous stage to obtain more features.The experimental results show that the PSNR values of the proposed network model reach 32.95 dB and 31.74 dB in Set12 and BSD68 datasets,re-spectively.At the same time,the subjective visual effects such as image edges and other details are recovered better and clearer.
NETL
NSTL
万方数据
