Super-Resolving Face Image by Facial Parsing Information

Face super-resolution is a technology that transforms low-resolution face image into the corresponding high-resolution one. In this paper, we build novel parsing map guided network which extracts prior (i.e., map) directly from for following utilization. To exploit extracted fully, attention fusion block carefully designed, can not only effectively explore information of map, but also combines powerful mechanism. Moreover, in light features contain more precise spatial while provide strong contextual information, hope to maintain and utilize these complementary information. achieve goal, develop multi-scale refine take advantage feature representations. Experimental results demonstrate our method outperforms state-of-the-arts terms quantitative metrics visual quality. The source codes will be available at.