Single Image Super-resolution Model Based on Improved Sub-pixel Convolutional Neural Network

Abstract

In order to improve the clarity of a single image after super-resolution and reduce the complexity of calculation, this paper uses a neural network model based on sub-pixel convolution to speed up the image processing speed and make the image details after super-resolution more clear. In the feature extraction, the image features are first extracted by using a smaller convolution kernel, then the image features are enlarged through the up-sampling process, and finally the feature extraction is performed through the convolution operation again. At the same time, in order to better preserve the image features, this paper also adds a feature compensation module. When the magnification is 3, the PSNR value is higher than the ESPCN (+1.03db). The sub-pixel convolutional network model in this paper effectively reduces the computational complexity and improves the image quality, and provides an idea for the optimization of single image super-resolution.