The respected Comrade
"Great efforts should be made in the field of cutting-edge science and technology to develop technologies with a world-class competitive edge."
As a typical field in image processing technology, super-resolution imaging technology is a technique for obtaining high-quality images from low-resolution images without the need for expensive scanners, and plays an important role as a pre-processing for image processing and editing, analysis and recognition of medical and satellite images, for image communication and camera monitoring systems to increase the accuracy of recognition and to provide better visual effects to the user. Recently, deep learning techniques have been widely applied for image processing and recognition, and deep learning has been actively studied for super-resolution single-image processing, and many achievements have been made.
In the super-resolution of single images by deep learning, research is being conducted to improve the training speed of the network, simplify the network structure, reduce the computational effort and realize super-resolution on any scale.
In this paper, we propose a single-image super-resolution method using cross-residual network and wavelet transform to simplify the structure of deep neural network, speed up training and realize multi-scale super-resolution. To further increase the information exchange between different layers and to improve the network performance by more efficient learning, we propose a cross-residual block, propose a structure of cross-residual network to realize super-resolution end-to-end mode, and define a loss function for network learning.
A cascade and learning approach of the cross-residual network is proposed to achieve multi-scale super-resolution. The effectiveness of the proposed model is verified by comparing the test images with previous models.
The results of the above study were published in the SCI International Journal of "Wavelet Multiresolution and Information Processing" under the title of "Single image super-resolution based on cross-residual network and wavelet transform" (https://doi.org/10.1142/S0219691323500169).