Image Super-Resolution Reconstruction Based on Deep Residual Network

In recent years, image super-resolution reconstruction methods based on convolution neural network have developed rapidly. In order to solve the problems of large amount of computation and long training time caused by increasing the depth and width of the network, this paper proposes an image super-resolution reconstruction model 1 based on deep residual network. The structure of the model is simple, it includes feature extraction block, stacked feature enhancement block and reconstruction block. The feature enhancement block is composed of an enhancement unit and a non-linear unit. The enhancement unit extract different kinds of features through convolution kernels in different paths, and the non-linear unit is responsible for extracting various non-linear features. On this basis, a model 2 is also proposed to add the connection between blocks. The experimental results of four data sets such as Set5 show that the proposed two model can speed up the convergence of the model, enhance the features better, thereby, improving the quality of image reconstruction.