Image Super-Resolution Reconstruction Based on Big Data and Cloud Computing

Abstract

Image super-resolution reconstruction can reconstruct low-resolution images into high-resolution images, which is an important application of big data combined with cloud computing. Using big data technology can mine the useful information of a large number of images, and cloud computing can reduce the model computation. However, existing super-resolution models are difficult to train and have problems such as artifacts, blurred detail texture and too smooth after image reconstruction. To solve the above problems, we propose the Multi-scale double Attention mechanism based on Residual Dense Generative Adversarial Network (MARDGAN), which uses multi-branch paths to extract image features of different scale sizes, to obtain multi-scale features information. We also design the double attention mechanism block (CSAB) and combine it with the Enhanced Residual Dense Block (ERDB) to form the deep residual dense attention module (DRDAM) to extract multi-level depth feature information. The perceptual capability of the model is improved by adding pixel loss, perceptual loss, and adversarial loss. The experimental results show that our proposed MARDGAN has shorter training time. And it can use the original image information more effectively than other methods on multiple benchmark datasets to recover super-resolution images with clearer details and better realism.