基于残差密集注意力网络的图像超分辨率重建

Image super-resolution reconstruction based on residual dense attention networks

针对现有图像超分辨率重建算法中细节丢失和图像边缘模糊等问题,提出了一种基于残差密集注意力网络的图像超分辨率重建方法。该方法采用了密集连接和残差连接的结构来构建残差网络,充分利用低层特征与高层特征之间的信息交互,提取更高层次的图像特征。同时,融合通道注意力和空间注意力自适应地选择重要特征,并将这些特征进行加权融合,从而更好地恢复图片的纹理细节。实验结果表明,文中所提方法在峰值信噪比(PSNR)和结构相似度(SSIM)上表现优异。

An image super-resolution reconstruction method based on residual dense attention networks is pro-posed to address the problems of detail loss and blurred image edges in existing image super-resolution recon-struction algorithms.The method employs a structure of dense connections and residual connections to construct the residual network,making full use of the information interaction between low-level features and high-level features to extract higher-level image features,Meanwhile,fused channel attention and spatial attention adaptive-ly select important features and weighted fusion of these features,thus better recovering the texture details of the image.Experimental results show that our proposed method performs well in terms of peak signal-to-noise ratio(PSNR)and structural similarity(SSIM).