Single image super-resolution method based on reversible network

Abstract

The invention discloses a single image super-resolution method based on a reversible network, and belongs to the field of image processing. According to the method, a network structure of a super-resolution model is constructed by introducing a reversible network; mutual mapping of a high-resolution image space and a low-resolution image space is realized by utilizing the reversible property of areversible network; the super-resolution process is optimized in the low-resolution direction and the high-resolution direction, the problem that other super-resolution methods based on deep learningcannot effectively utilize the mutual dependence between high-resolution images and low-resolution images is solved, and therefore the image super-resolution capability of the model is improved. A weight matrix of the 1 * 1 reversible convolution layer is initialized by introducing singular value decomposition, so that the propagation speed of the inverse process of the 1 * 1 reversible convolution layer is increased; by the adoption of the method, the super-resolution process of a single image can be effectively achieved, and the super-resolution image with good texture details and visual effects is generated through the low-resolution image.

Description

technical field [0001] The invention relates to a single image super-resolution method based on a reversible network, belonging to the field of image processing. Background technique [0002] Image Super Resolution refers to recovering a high-resolution image from a low-resolution image or image sequence. Single image super resolution (SISR) is to establish a mapping between low-resolution images and high-resolution images, and generate super-resolution images from input low-resolution images. Current super-resolution methods mainly include interpolation-based, reconstruction-based, and learning-based methods. Common interpolation methods include bilinear interpolation, bicubic interpolation methods, etc., but the reconstructed image obtained by interpolation is prone to blur, jagged, and lacks texture details. The reconstruction-based method uses the low-resolution image as a constraint and combines the prior knowledge of the image to reconstruct and restore, such as iter...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the problem that the current existing technology does not utilize the interdependence between the low-resolution image and the high-resolution image, and cannot use the mutual information between the two images more effectively, thus affecting the effect of model image super-resolution, The present invention provides a single image super-resolution method based on a reversible network. The method uses a reversible network to construct a network model for super-resolution, and then inputs a low-resolution image to one end of the network model to generate a super-resolution image. Input the high-resolution image to the other end of the network model to generate a low-resolution reconstruction image, and use the generated super-resolution image and the difference between the low-resolution reconstruction image and the real high-resolution image and low-resolution image to design The optimized objective function updates the parameters of the network model by minimizing the value of the objective function to improve the super-resolution capability of the network model

Images