Multi-level Feature Fusion Mechanism for Single Image Super-Resolution

Convolution neural network (CNN) has been widely used in Single Image Super Resolution (SISR). As the network deepens, the learning ability of the network becomes more and more powerful. However, most SISR methods based on CNN do not make full use of the hierarchical features and the learning ability of the network. These features cannot be extracted directly by subsequent layers, so the previous layer hierarchical information has little impact on the output and performance of subsequent layers relatively poor. To solve the above problem, a novel Multi-Level Feature Fusion network (MLRN) is proposed, which can make full use of global intermediate features. I also introduce Feature Skip Fusion Block (FSFblock) as a basic module. Each block can be extracted directly to the raw multi-scale feature and fusion multi-level feature, then learn spatial correlation. The correlation among the features of the holistic approach leads to a continuous global memory of information mechanism. Experiments on public datasets show that the method MLRN can be implemented, which has a favorable performance than most of the current methods.