Residual Dense Network for Image Super-Resolution

A very deep convolutional neural network (CNN) has recently achieved greatsuccess for image super-resolution (SR) and offered hierarchical features aswell. However, most deep CNN based SR models do not make full use of thehierarchical features from the original low-resolution (LR) images, therebyachieving relatively-low performance. In this paper, we propose a novelresidual dense network (RDN) to address this problem in image SR. We fullyexploit the hierarchical features from all the convolutional layers.Specifically, we propose residual dense block (RDB) to extract abundant localfeatures via dense connected convolutional layers. RDB further allows directconnections from the state of preceding RDB to all the layers of current RDB,leading to a contiguous memory (CM) mechanism. Local feature fusion in RDB isthen used to adaptively learn more effective features from preceding andcurrent local features and stabilizes the training of wider network. Afterfully obtaining dense local features, we use global feature fusion to jointlyand adaptively learn global hierarchical features in a holistic way. Extensiveexperiments on benchmark datasets with different degradation models show thatour RDN achieves favorable performance against state-of-the-art methods.