High-efficiency infrared image super-resolution algorithm based on a cascaded deep network

Abstract

Single image super-resolution (SISR) is a notoriously challenging ill-posed problem that aims to obtain a high-resolution output from one of its low-resolution versions. Recently, powerful deep learning algorithms have been applied to SISR and have achieved state-of-the-art performance. In this paper, a high-efficiency infrared image super-resolution algorithm based on a cascaded deep network is proposed. In this method, the low-resolution infrared image is directly processed without the preprocessing of bicubic interpolation up-sampling that can reduce the complexity of the network and the amount of computation. The network structure consists of two layers of the network. The sub-pixel convolution in each layer can enlarge the image size by twice and make the input image size to reach the final high-resolution image size. Besides, we utilize multi-scale feature extraction blocks to extract features from the same feature image by using multiple convolution kernels of different sizes, which makes the feature image information more abundant. The experimental results show that the test speed of each image in our network is 0.046 seconds, which manifests our proposed algorithm has high efficiency of infrared image super-resolution.