MCFPN: Multi-Path Cross Fusion Pyramid-Like Network for Image Super-Resolution Reconstruction

In the multipath reconstruction process, the currently existing super-resolution reconstruction methods ignore the structural similarity between features, which makes high-frequency features, such as texture detail features, difficult to be extracted. To solve this problem, a multi-path cross fusion pyramid-like network (MCFPN) is proposed for image super-resolution reconstruction in this paper. First, a shuffle separable attention (SSA) block is designed to select cross-channel feature information which improves the network's ability. Second, a multi-path cross fusion pyramid-like structure is constructed to mine the hierarchical features under different dimensions. Finally, a forward and reverse cross fusion strategy is designed to efficiently utilize structural information in different dimensions. Experimental results show that MCPFN is well adapted to the reconstruction tasks.