Multi-branch network and method for motion blur super-resolution

Abstract

The invention discloses a multi-branch network and method for motion blur super-resolution, wherein the network comprises a deblurring module, a fusion module, a reconstruction module, an image outputmodule which are connected successively, and also comprises an SR feature extraction module which is connected with the deblurring module in parallel. The method comprises the steps: acquiring an image and processing the image; establishing a data set; constructing a network structure; and calculating and outputting processing results. A network comprising three branches is proposed, high-frequency deblurring and low-frequency deblurring are respectively realized by the first two branch; a dual double-branch generative adversarial network is formed as a deblurring module; high-frequency information deblurring is used as a single branch, so that the restoration of detail parts of the image is further emphasized; the third branch realizes super-resolution feature extraction, deblurring andsuper-resolution features are fused through the fusion module, then reconstruction is carried out, effective super-resolution treatment is carried out on a motion blurred image caused by relative motion between a camera and a scenery, and a pleasant high-resolution image is generated.

Description

technical field [0001] The invention relates to the field of image processing, in particular to a multi-branch network and method for performing motion blur super-resolution. Background technique [0002] Super-resolution aims to recover high-resolution (HR) images from low-resolution (LR) images, which has received a lot of attention and progress in recent years. Super-resolution methods are used to input inherently low-resolution images, such as those from surveillance cameras and moving cameras, to generate pleasing high-resolution images that can significantly improve the performance of other machine vision tasks . In practical application, due to the motion of the camera or the target, the captured image tends to become blurred, and this blurring is called motion blur. At this time, the super-resolution method will amplify the motion blur around the moving person and generate blurred high-resolution patches, which cannot obtain normal high-resolution images. Therefore...

AI Technical Summary

Problems solved by technology

[0005] The present invention mainly solves the technical problem that the original super-resolution and deblurring have an irreversible impact on the image, provides a multi-branch network and method for motion blur super-resolution, and proposes a network including three branches, the first two respectively realize high-frequency De-blurring and low-frequency de-blurring constitute a dual-branch generative confrontation network as a de-blurring module. High-frequency information de-blurring is used as a single branch to emphasize the restoration of image details. The third branch realizes super-resolution feature extraction. Through the fusion module, the deblurring and super-resolution features are fused and reconstructed, and the motion blurred image caused by the relative motion between the camera and the scene is effectively super-resolved to generate a pleasing high-resolution image

Images