Title：Video blind denoising method and device based on deep learning
Country：China
Patent No.：202010294520.3
Legal Status：Authorized
Inventor：Xiang Xie, Shaofeng Zou, Guolin Li, Songping Mai, Zhihua Wang
Assignee：International Graduate School at Shenzhen, Tsinghua University; Tsinghua University
Address：Tsinghua Campus, Xili University Town, Nanshan District, Shenzhen City 518055, Guangdong Province
Filing Date：2020-04-15
Issue Date：2023-04-14
Abstract：
The invention provides a video blind denoising method and device based on deep learning. The method comprises the following steps of: taking a video sequence containing a preset number of frames froma to-be-denoised video, taking an intermediate frame of the video sequence as a noisy reference frame, and carrying out optical flow estimation on the noisy reference frame and each of other frames ofimages in the video sequence to obtain an optical flow field between a plurality of two-frame images; converting each of other frames of images in the video sequence into the noisy reference frame for registration according to the optical flow field to obtain multiple frames of noisy registration images; constructing a denoising network based on a convolutional neural network, taking the multipleframes of noisy registration images as network input, adopting the noisy reference frame as a reference image of the network, and conducting frame-by-frame iterative training and denoising by using anose2noise method to obtain a denoised image corresponding to the noisy reference frame. According to the scheme, blind denoising of the video can be realized only by using a single video without acquiring a large amount of noisy data, clean data and an accurate noise distribution model.
Patent Certificate：
PDF/Jpg