Research On Single Image Super-resolution Reconstruction Algorithm Based On Deep Learning

Image resolution typically characterizes the ability of a display system or a image to convey detailed information.In general,the higher the resolution of an image,the richer the detail of the image,and the higher the quality of the image.Image super-resolution reconstruction algorithm is an effective algorithm to improve image quality,which can increase the spatial resolution of images and enhance the details of images.The image super-resolution algorithms based on deep learning have many advantages.First,there are a large number of images on the Internet,which is sufficient for deep learning.Subsequently,the learned model can directly learn the reconstructed super-resolution image,and the calculation is simple and straightforward.Finally,the mathematical model based on deep learning has a strong ability to express high-level features and represent the structure of the image.In this thesis,we analyze the characteristics of image super-resolution reconstruction tasks,an extremely deep residual network structure is designed to reconstruct the image super-resolution.The residual module in the network structure uses the channel attention mechanism and the coordinate convolution technique,and the 6 residual modules form a residual module group through skip connections,and finally 32 residual module group form a residual structure by using a long jump connection.The corresponding experiments have proved the validity of the structure.After analyzing the properties of the loss function commonly used in super-resolution algorithms,the super-resolution is improved by combining the content loss function and the texture detail loss function.In the end,it achieved good results on the public data set..Finally,we combine relativistic generative adversarial networks and convolutional neural network to make the reconstruction results have more high-frequency details and the effect is more realistic.