Image Super-resolution Reconstruction Method Based on Residual Subpixel Convolutional Network

Image Super-resolution Reconstruction Method Based on Residual Subpixel Convolutional Network

Image super-resolution reconstruction is a mothed to generate clear images from vague images. Due to the image reconstructed by depth learning method has better display effect. Therefore, this paper proposes an image super-resolution reconstruction method based on residual sub-pixel convolutional network. Firstly, aiming at the problem that adding residual network will reduce the network efficiency, the batch normalization layer is deleted in the residual structure to increase the model efficiency. Then, so as to fully utilize the feature details of the vague image, the multi-layer feature image information is extracted and filtered through the jump connection and global feature multiplexing module. The network can take advantage of image details with different depths. Finally, so as to increase the speed of network reconstruction, the sub-pixel convolution network enlarges and rearranges the obtained feature images to obtain clear images. It can be concluded from the experimental results that this method obtains better subjective visual effect, and the experimental data of objective evaluation standard peak signal to noise ratio (PSNR) and Structural similarity (SSIM) are also larger than the classical bicubic interpolation method, super-resolution convolutional neural network (SRCNN) algorithm and super-resolution using very deep convolutional network (VDSR) algorithm show advantages of this method.