Image Super-resolution Algorithm Based on Dual Regression Networks with Cross-scale Non-local Attention Mechanisms
Guanxing Li, Yanjun Wei, Tianping Li, Meng Li · 2022 International Conference on Machine Learning and Intelligent Systems Engineering (MLISE) · 2022
Image super-resolution reconstruction is an ill-posed problem in nature because there are infinitely many high-resolution images that can be reconstructed from a low-resolution image. To limit the solution space and make good use of the widely existing cross-scale feature similarities in natural images, we propose a dual regression network based on the cross-scale non-local attention mechanism, which can not only restrict the solution space of the image super-resolution but also make the network better embrace the abundant external information. The method has been tested on five benchmark datasets and both achieve good results.