中国科学院云南天文台机构知识库

针对现有单图像超分辨率重建时主要采用简单链式堆叠的单一网络存在着的层间联系弱、网络关注点单一以及分层特征不能充分利用等问题，提出了一种复合的深度神经网络用于提升图像超分辨重建性能。该方法首先使用特征提取层提取低分辨率图像的初始特征；再将初始特征分别送入两个子网络，一个子网络负责图像细节的提取与运算，另一子网络负责图像噪声降解与消除；然后将两个子网络输出的深层次抽象特征与初始特征相结合，最后通过重建层获得超分辨率图像。以峰值信噪比与结构相似性为评价指标，在Set14测试集上使用放大因子3进行实验，将复合网络与算法Bicubic、SelfEx、SRCNN、VDSR和RED等进行对比，实验结果发现，PSNR分别提高了2.27 dB、0.66 dB、0. 54 dB、0.05 dB和0.21 dB，而SSIM则分别提高了6.08、1.54、1.41、0.36和0.09个百分点。 

Aiming at the problems of weak inter-layer connection, single network focus and insufficient utilization of hierarchical features in existing single image super-resolution reconstruction using simple chain stack, a composite depth neural network is proposed to improve the performance of image super-resolution reconstruction. Firstly, feature extraction layer is used to extract the initial features of low-resolution images; then, the initial features are fed into two sub-networks, one is responsible for extracting and combining image details, the other is responsible for image noise degradation and elimination; and then, the deep abstract features and initial features output by the two sub-networks are fed into two sub-networks respectively. Finally, the super-resolution image is obtained by feature combination and reconstruction. Taking peak signal-to-noise ratio and structural similarity as evaluation indexes, the experiment was carried out by using amplification factor 3 in Set14 test set. The experimental results showed that PSNR increased by 2.27 dB, 0.66 dB, 0.54 dB, 0.05 dB and 0.21 dB respectively, while SSIM improved by comparing the combined network with algorithms Bicubic, SelfEx, SRCNN, VDSR and RED. 6.08, 1.54, 1.41, 0.36 and 0.09 percentage points.