Lightweight Image Super-Resolution Reconstruction Algorithm Based on Spectral Norm Regularization GAN and ShuffleNet

Abstract

In response to the low feature extraction ability of current image super-resolution models, an image reconstruction algorithm with an improved generative adv.9ersarial network is proposed. On the basis of the image super-resolution algorithm based on the generated adversarial network, the spectral norm and least square relative discriminator are introduced, and then the latest version ShuffleNetV is added to improve the accuracy of the model. A lightweight image super-resolution reconstruction algorithm based on improved generative adversarial network is tested. The test results show that the evaluation scores for WOMAN images, HEAD images, BUTTERFLY images, and BABY images using the research method were 43.6, 33.8, 27.9, and 46.3, respectively. The K values of the reconstructed samples in the Set5 dataset were mainly concentrated in the range of 0.4008.5, while in the Set14 dataset, the K values were roughly distributed in the range of 0.20 to 1.10. In the ablation experiment, the PI value of the research model is 2.11, indicating that the research model can generate high-quality images that are closest to the real high-resolution images in terms of perceptual features and texture details. From this, lightweight image super-resolution reconstruction algorithms with improved generative adversarial networks have significant performance advantages, which can promote technological progress in image super-resolution reconstruction.