基于注意力机制与特征融合的图像超分辨率重建

目前多数利用卷积神经网络进行图像超分辨率重建的方法忽视对自然图像固有属性的捕捉，并且仅在单一尺度下提取特征。针对该问题，提出一种基于注意力机制和多尺度特征融合的网络结构。利用注意力机制融合图像的非局部信息和二阶特征，提高网络的特征表达能力，同时使用不同尺度的卷积核提取图像的不同尺度信息，以保存多尺度完整的信息特征。实验结果表明，该方法重建图像的客观评价指标和视觉效果均优于Bicubic、SRCNN、SCN和LapSRN方法。

Image Super-Resolution Reconstruction Based on Attention Mechanism and Feature Fusion

In super-resolution image reconstruction,most of the existing methods using Convolutional Neural Network(CNN)neglect the inherent attributes of natural images,and extract features only at a single scale.To address the problem,this paper proposes a network structure based on attention mechanism and multi-scale feature fusion.By using the attention mechanism,the non-local information and second-order features of the image are fused to improve the feature expression ability of the network.At the same time,different scales of convolutional kernels are used to extract different scales of information of the image,so as to preserve the complete information characteristics at different scales.Experimental results show that the reconstructed image by the proposed method outperforms Bicubic,SRCNN,SCN and LapSRN methods in terms of objective evaluation metrics and visual quality.