An Optimization of Image Super-Resolution Method Based on Residual Network

Abstract

In this paper, an optimization with deep residual networks for the super-resolution method will be proposed. Meanwhile, the optimized method will be used into MNIST database to improve the image resolution, and the accuracy of the image identification of the improved database will be test by the residual network. In the optimization design, there are 4 convolutional layers, and 5 residual blocks will be used to replace the 3 convolutional layers of initial design, and there are 3 convolutional layers in each residual block. After the residual network testing, the experiment is positive and effective, and total number of errors is 34, and the accuracy is 99.43%.