Images super-resolution using improved generative adversarial networks

Image super-resolution (ISR) is an important image processing technology to improve image resolution in computer vision tasks. The purpose of this paper is to study the super-resolution reconstruction of single image based on the depth learning method. Aiming at the problem that the existing pixel loss-based super-resolution image reconstruction algorithms have poor reconstruction effect on high-frequency details, such as textures, a lighter algorithm is proposed on the basis of the existing deep learning method (SRGAN). Firstly, the feedback structure is applied in the generator to process the feedback information and enhance the high frequency information of the image. Secondly, a general residual feature aggregation framework (RFA), is applied to make full use of the residual information of each layer to improve the quality of the SR image. Finally, the solution space of the function is further reduced and the image quality is improved by using a new loss function. The algorithms are implemented on pytorch framework. The experimental results on VOC2012 data sets show that, compared with the original SRGAN algorithm, the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) of the proposed algorithm on the benchmark data set Set5 are improved by 0.83dB and 0.028, respectively, on Set14, the PSNR and SSIM of the proposed algorithm are improved by 0.56dB and 0.009, on Urban100, the PSNR and SSIM of the proposed algorithm are improved by 0.51dB and 0.031, on BSD100, the PSNR and SSIM of the proposed algorithm are improved by 0.33dB and 0.014, and compared with other improved algorithms, the effect of this algorithm is also better than other algorithms.