Hierarchical Recursive Network for Single Image Super Resolution.

Minglan Su,Shenqi Lai,Zhenhua Chai,Xiaoming Wei,Yong Liu
DOI: https://doi.org/10.1109/icmew.2019.00108
2019-01-01
Abstract:Super Resolution (SR) technique aims to reconstruct the high resolution (HR) image from the observed low resolution (LR) one, which is a significant application in our daily life. In this paper, we propose a novel structure named hierarchical recursive network (HRN), which consists of several sub networks and will reconstruct the HR progressively. In each sub network, the LR feature map will be used as input, the contextual information will be explored and the predicted residuals together with the transposed convolutional outputs will be fused to the finer one. Besides, our network can generate multi-scale HR images with a single model and thus is potentially useful in practical applications. Extensive experimental results show that our proposed method can achieve the state-of-the-art performance.
What problem does this paper attempt to address?