金字塔方差池化网络的图像超分辨率重建OA北大核心CSTPCD

To reduce the impact of high-frequency information loss on image reconstruction and further enhance the mining of feature information,a generation network is constructed with the pyramid variance pooling module as the core.Firstly,the network uses different levels of variance pooling to mine the feature information contained in low-resolution images,and combines the pyramid structure to obtain the context information of different scales and different sub-regions,so as to further enrich the amount of feature information.Then,the dense connection structure is used to enhance the correlation of feature information to improve the expressive ability of the network.Finally,the group normalization operation is introduced to strengthen the convergence of the network.The experimental results show that compared with other methods on the open test sets of Set5,Set14,and DIV2K100,the peak signal-to-noise ratio increases by an average of 0.509 dB and the structural similarity increases by an average of 0.016 when the amplification factor is 4.The proposed model not only improves the peak signal-to-noise ratio and structural similarity to a certain extent,but also has more realistic details in the visual effect of the reconstructed image.