Super-Resolution Reconstruction of Single Image based on Multiscale Recursive Residual Network
Mi Dong, Jian Huang · IOP Conference Series Materials Science and Engineering · 2020
Abstract In order to solve the problems of single image super-resolution algorithm based on convolutional neural network, such as shallow network structure, single feature extraction scale and fuzzy texture of reconstructed image, a single image super-resolution reconstruction method based on multi-scale convolutional neural network is proposed. The convolution of several different scales is used to check the original low-resolution image for feature detail extraction, and the recurrent residual network is used to gradually restore the high-frequency information of the image with fewer parameters. Finally, the final reconstructed image is obtained by fusion of the features extracted by different convolutional kernels. The experimental results show that the algorithm proposed in this paper has obtained better image super-resolution result, can obtain more detailed information, and get better visual effect. The classical reconstruction methods such as bicubic, SRCNN, FSRCNN, ESPCN, VDSR are compared, and the peak signal-to-noise ratio (PSNR) and the structure similarity (SSIM) are higher than that of the existing algorithms.