The invention discloses a video restoration model training method based on a deep network and a video restoration method

Abstract

The invention discloses a video restoration model training method based on a deep network and a video restoration method. The method comprises the following steps: obtaining a plurality of frames of clear images from a standard clear video; Carrying out Gaussian blur processing on each frame of clear image to obtain a blurred image corresponding to each frame of clear image; Taking a {clear imagesequence, fuzzy image sequence} pair consisting of continuous n frames of clear images and corresponding n frames of fuzzy images as a training sample, thereby obtaining a training set consisting of all training samples; Establishing a video restoration model that is connected by n-1 codec networks in turn and used for image restoration of blurred image In based on blurred image In and its previous n-1 frame blurred images In-1-I1; And training the video restoration model by using the training set so as to obtain a target video restoration model. According to the invention, the restoration efficiency and restoration effect of the blurred video can be improved.

Description

technical field [0001] The invention belongs to the technical field of image processing, and more specifically relates to a deep network-based video restoration model training method and a video restoration method. Background technique [0002] With the rapid development of informatization, the use of electronic products such as smart phones and VR equipment is becoming more and more popular. At the same time, technologies such as video surveillance and video transmission are more and more widely used. In the process of video shooting or transmission, it is easy to cause loss or distortion of video information due to various reasons, resulting in blurred video. [0003] Image restoration has always been an important research direction in the field of image processing. Early image restoration techniques divide image restoration into non-blind image restoration and blind image restoration according to whether the point spread function (PSF) of the blurred image is known. Bec...

AI Technical Summary

Problems solved by technology

The patent "A method for removing non-uniform motion blur in images based on deep neural network" (application number: CN 104680491) proposes to use a deep convolutional network to estimate the point spread function of blurred image blocks, and then use the optimized Markov random field The model obtains the pixel-by-pixel point spread function of the blurred image. Finally, based on the estimated point spread function, a clear image is obtained using the image non-blind restoration algorithm. This method successfully introduces deep learning into image restoration, but this method needs to pass through the deep network first. Obtain the point spread function of the blurred image, and then use the non-blind rolling algorithm to restore the blurred image, which cannot achieve end-to-end image restoration, and it is more complicated to use in practical applications

[0005] Generally speaking, the existing blurred image restoration methods have certain limitations, which makes the video restoration method based on blurred image restoration also have the problems of low efficiency and poor effect.

Images