Lightweight image super-resolution with a feature-refined network

In recent years, deep Convolutional Neural Networks (CNNs) have achieved impressive successes on the Single Image Super-Resolution task (SISR). However, it remains difficult to apply these CNN-based SISR methods in embedded devices due their high memory and computational requirements. To alleviate this issue, we focus lightweight methods. The observed similarity between feature maps CNNs serves as inspiration explore design of a cost-efficient module obtain whose representation ability is roughly equivalent that conventional convolutional layer. We thus propose shadow applying simple linear transformations with lower cost generate similar maps. Based module, Feature-Refined Block (FRB) learn more representative features. Besides, Global Dense Feature Fusion (GDFF) structure construct Network (FRN) such FRBs for SISR. Extensive experimental results demonstrate superior performance proposed FRN comparison state-of-the-art methods, while consuming relatively low computation resources.