Real image denoising method based on generative adversarial network noise modeling

Abstract

The invention discloses a real image denoising method based on generative adversarial network noise modeling, and aims to solve the problem that real image noise cannot be effectively removed due to difficulty in explicit modeling by utilizing a deep convolutional neural network. The method comprises the following steps: firstly, learning real image noise distribution by utilizing a generative adversarial network, and constructing a paired data set; and secondly, training a denoising network by using the constructed data set to realize removal of real noise. According to the method, the denoising network is established by using the residual block, and the noise estimation algorithm and the reversible up-down sampling operation are introduced into the denoising network, so that the blind image denoising performance of the single network model is improved, and the edge information and detail features of the original image are reserved as much as possible while unknown distribution noiseis effectively removed.

Description

technical field [0001] The invention relates to the technical fields of computer vision, computer application, etc., and in particular to a real image denoising method based on generative anti-network noise modeling. Background technique [0002] Images are often degraded due to the interference of various noises during the acquisition or transmission process, which will have an adverse effect on the visual effect of the image and subsequent processing. Therefore, denoising preprocessing must be performed on the image to suppress noise and improve image quality. Existing image denoising methods can be mainly divided into model-based optimization methods and discriminative learning-based methods. The model-based optimization method starts from the Bayesian point of view. When the likelihood probability of the image is known, the prior information of the image is used to restore the image. Among them, the BM3D algorithm and the WNNM algorithm are superior. The BM3D algorithm ...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the problem that the noise distribution is complex and difficult to model explicitly, the present invention proposes a real image denoising method based on generative adversarial network noise modeling, and at the same time introduces a noise estimation module into the network to improve the effect of blind denoising, so that The constructed network can effectively remove the real noise while retaining the original details and edge information

Images