Research On Image Super-Resolution Reconstruction Algorithm Based On Deep Residual Network

In today's society,information is mainly transmitted in the form of digital images.Due to the limitations of the imaging equipment and the impact of environmental factors,in many cases,only low-resolution images can be obtained,so the use of image super-resolution reconstruction technology can obtain clear images to meet actual needs.Image super-resolution reconstruction technology uses software algorithms to reconstruct low-resolution images into high-resolution images.In recent years,deep learning-based image super-resolution reconstruction algorithms have emerged endlessly and become a research hotspot.This thesis mainly studies the image super-resolution reconstruction algorithm based on the deep residual network.The main research work is as follows:1.Research the image super-resolution reconstruction algorithm based on deep residual network,summarize the characteristics of most models,and form a conventional model of super-resolution reconstruction based on deep residual network,which mainly includes feature extraction block,feature enhancement block and reconstruction block.Aiming at the problem of limited feature enhancement capabilities of feature enhancement blocks in conventional models,an improved model 1 based on deep residual network is proposed.This model mainly improves the feature enhancement block,each feature enhancement block consists of an enhancement unit and a non-linearization unit.The enhancement unit extracts different kinds of features through convolution kernels under different paths.The non-linearization unit is responsible for extracting various non-linear features and enhancing the representation ability of the features.2.Aiming at the problem of weak connection between blocks in the improved model 1,this thesis proposes improved model 2 based on the idea of dense connection,and increases the connection between blocks in the model based on the improved model 1.The enhancement unit of each feature enhancement block fuse the feature information extracted by the previous feature extraction block,so that the features extracted from the earlier layer are fused into the deeper layer to enhance feature reuse.3.Perform reconstruction experiments on the above two models on 4 commonly used data sets such as Set5,Set14,BSD100 and Urban100.Experimental results prove that compared with the classic reconstruction model based on the deep residual network,the improved feature enhancement block in the improved model 1 proposed in this thesiscan more effectively enhance the feature representation ability and improve the image reconstruction effect;the improved model 2 with the connection between blocks can enhance the feature transfer,effectively accelerate the network convergence,and further improve the image reconstruction effect.