A super-resolution reconstruction algorithm based on feature fusion

Super-resolution reconstruction is a technique for recovering corresponding high-resolution images from low-resolution images. Aiming at the problem of insufficient learning ability of the generative network in the SRGAN, a super-resolution reconstruction algorithm of generative adversarial network based on feature fusion is proposed. Recursive residual networks and prior knowledge are used to extract edge and texture features, while densely connected networks are used to fully utilize the features. Aiming at the problem that the input of the discriminative network of the SRGAN has a high computational complexity for the entire image, a method for discriminant residual discrimination is proposed. Two difference maps are used as input images of the discriminative network, and two discriminative methods are used for training. The test data set was used to verify the performance of the algorithm. The subjective index was significantly better than the original algorithm. The objective index PSNR increased by 0.77db, and SSIM increased by 2.8%.