Image Super-Resolution Algorithm Based on Multi-Scale Convolution Neural Network

The single image super-resolution problem is a typical image inverse problem. In recent years, deep learning is widely used in image super-resolution. In order to improve the performance of super-resolution algorithms, in this paper, the ideas of multi-scale and residual training are utilized. An image super-resolution algorithm that exploits the multi-scale convolution neural network is proposed. This algorithm exploits the multi-scale convolution kernels and the shrinkage-extension structure to extract image multi-scale information. Skip connection is used in the network structure to improve the quality of image reconstruction，which can transmit information effectively. Moreover, it can compensate for the loss of image details resulting from the use of down-sampling and up-sampling. Compared with other algorithms, the experiments shows that our algorithm can not only achieve better performance, but also has the faster convergence speed.