Video deblurring method, device, device and storage medium

Abstract

The present application provides a video deblurring method, device, equipment, and storage medium, which can be applied to various scenarios such as cloud technology, artificial intelligence, intelligent transportation, assisted driving, and video. The method includes: acquiring the current image in the video, and the N neighboring images of the current image; extracting the first feature information of the current image at M scales, and the second feature information of each of the N neighboring images at M scales; for the N neighboring images For each adjacent image of , align the second feature information of the adjacent image at M scales with the first feature information at M scales respectively; according to each of the N adjacent images aligned at M scales The second feature information is to obtain the current image after deblurring. That is, in the feature domain, the present application aligns the first feature information of the current image with the second feature information of adjacent images, thereby improving the video deblurring effect.

Description

technical field [0001] The embodiments of the present application relate to the technical field of image processing, and in particular, to a video deblurring method, device, device, and storage medium. Background technique [0002] In the process of video capture, due to the shaking of the camera equipment, the rapid movement of objects and other reasons, the video will inevitably be blurred. In order to improve the display effect of the video or improve the accuracy of video-based task analysis, it is necessary to deblur the blurred video, that is, restore a clear video from the blurred video. [0003] Current video deblurring methods use an optical flow model to align the current image with its neighbors in the image domain. Based on the aligned current image and adjacent images, the attention mechanism is used to calculate the regional similarity. The features of the similar area are fused to the current area to obtain the fused features of the current image, and finall...

AI Technical Summary

Problems solved by technology

[0004] However, when using the optical flow model to align the current image with adjacent images in the image domain, once the scene changes quickly and the distance is far away, the estimation of the optical flow field will be wrong, which will cause defects in the alignment results and lead to the current image’s inaccuracy. Poor deblurring effect

Images