Single-frame image super-resolution reconstruction method based on dual-channel feature migration network

Abstract

The invention discloses a single-frame image super-resolution reconstruction method based on a dual-channel feature migration network, and aims to solve the problems that detail information is lost due to increase of network depth of a deep super-resolution image reconstruction network and global-local texture similarity is not fully utilized by a local receptive field. An overall network is based on a residual mechanism, and a rear up-sampling module is used for carrying out spatial-scale up-sampling on an image to output a super-resolution reconstruction result y. According to the method, effective distribution of spatial repeated texture features can be kept, the event features are multiplexed by using a residual structure, the phenomenon that detail features disappear in the forward transmission process of the network is effectively prevented, and the super-resolution reconstruction quality of a single-frame image can be remarkably improved.

Description

technical field [0001] The invention belongs to the field of single-frame image super-resolution reconstruction, and in particular relates to a single-frame image super-resolution reconstruction method based on a dual-channel feature transfer network. Background technique [0002] The purpose of single-frame image super-resolution reconstruction (SISR) is to obtain a high-resolution image reconstruction output from a low-resolution image input, and to restore the details and texture information of the image to the greatest extent while increasing its spatial resolution. From the perspective of signal entropy, this is a process of entropy increase. In the case of no reference and no prior knowledge, this is an ill-conditioned problem of reconstructing the original real signal from the lossy degraded signal. From the early stage based on interpolation From the current learning-based method to the current learning-based method, since the human eye is not sensitive to high-frequ...

AI Technical Summary

Problems solved by technology

[0003] In view of the low reconstruction accuracy of the above reconstruction algorithms, insufficient use of feature information, and loss of texture details, etc., with the continuous development of deep learning in recent years, in order to optimize the quality of SISR problems and reduce the computational complexity of reconstruction, To improve the scene adaptation performance of the algorithm, the method based on deep neural network is also introduced into the SISR problem, but the simple neural network needs a large network depth to obtain better reconstruction performance, and the method based on convolution is only for The local receptive field of the image is effective, so the method of improving performance by deepening the network can only reconstruct the optimized result that conforms to the human visual perception. For some advanced pattern recognition tasks, its performance is often poor

Images