Super Resolution Reconstruction Method of Multi Feature Texture Mural Image Based on Twostagenet Model for Visual Communication
The super-resolution of a single image is of great value in image restoration.However,the complex compression artifacts and other problems hinder the development of this research.To solve this problem,we propose twostageNet;A network using a two-stage coarse to fine learning framework,which is mainly optimized for the super-resolution of compressed images.Specifically,TwostageNet consists of two main subnets:coarse and fine networks,in which recursion and residual learning are used respectively.The scale recursion framework is used to cooperate with the reconstruction.The framework can recursively reuse the higher coefficients of the filter for low scale factor learning.This will improve performance and parameter efficiency for higher factors.We trained two of our network versions to use different loss configurations to enhance complementary image quality.The proposed network is superior to the most advanced image and video super-resolution benchmark technology in both quality and quantity.The method is tested with the low resolution images simulated by smic-hs dataset.The experimental results show the effectiveness of the method.
NETL
NSTL
万方数据
