Image super-resolution reconstruction based on multi-scale learning and feature mapping network
super-resolution image reconstruction technology is aimed at traditional convolutional neural network reconstruction(Srcni)methods,which have the problems of low utilization of pixel features and weak recovery ability of high-frequency details,in this paper,a new method of multi-pixel recursive learning and feature mapping based on multi-scale convolution kernel and feature mapping network is proposed.By collecting low-resolution(LR)image data sets and image pixel features,based on SR image super-resolution reconstruction technology,using 1X1,3×3 scale convolution cores,the operations of dimensionality reduction,shallow feature extrac-tion,feature mapping and feature information fusion of image pixel data are made,and the feed-back results of local and global residuals after recursive learning are combined,the multi-scale low-resolution(LR)pixel features are mapped into the high-resolution(HR)pixel feature space to obtain the reconstructed super-resolution image after feature fusion.
万方数据
维普
