Progressive Face Super-Resolution Reconstruction Network Based on Relational Modeling

Abstract

Aiming at the imprecise details of the reconstructed face image caused by the large scale and ignoring the relationship modeling between different pixels in the upsampling process of most existing face super-resolution reconstruction algorithm models, a new progressive face super-resolution reconstruction network based on relationship modeling is proposed. The network mainly includes a detail information generation module based on progressive upsampling and a detail information enhancement module based on relational modeling. The step-by-step upsampling detail information generation module realizes the step-by-step generation of the face image detail information through the step-by-step upsampling operation. The detail information enhancement module based on relational modeling which adopts a linear and nonlinear relational modeling method optimizes the channel-level and spatial feature-level modeling of the detail information of the face image, and combines with the progressive upsampling detail information to achieve accurate reconstruction. Finally, through the experimental verification, the effectiveness of the algorithm proposed in this paper is proved.