A Densely Connected Face Super-Resolution Network Based on Attention Mechanism
Ying Liu, Zhanlong Dong, Keng Pang Lim, Nam Ling · 2020
Super resolution reconstruction of human face is a cost effective way to obtain high resolution images from its corresponding low resolution face. It is also known as face illusion. In order to obtain clearer texture details, this paper proposes a densely connected super-resolution algorithm based on attention mechanism which consists of feature extraction and image reconstruction. By integrating channel and spatial domain information of the feature map, the Multi Attention Domain Module (MADM) is proposed: Features are weighted and recombined by analyzing the relationship between channels and spatial information of feature maps. The features of different layers are fused using dense connections. Experimental results show that the proposed algorithm can improve by up to 0.5dB in PSNR and the reconstructed face image has clearer texture details compared to existing algorithms.