Image denoising method, system and device based on transfer learning and medium

Abstract

The invention discloses an image denoising method, system and device based on transfer learning, and a medium. The method comprises the steps: obtaining a to-be-denoised image; inputting the to-be-denoised image into a pre-trained denoising neural network based on transfer learning to be processed, wherein the denoising neural network based on transfer learning comprises a main denoising network and a noise distribution information extraction network; wherein the noise distribution information extraction network is used for extracting random noise distribution characteristics; preprocessing the random noise distribution characteristics, taking the preprocessed random noise distribution characteristics as dynamic normalization parameters of each residual module of the main noise reduction network, migrating the random noise distribution characteristics into data characteristics of the main noise reduction network, and accelerating the convergence rate of the main network; normalizing the image features extracted by each residual module of the main noise reduction network by using the dynamic normalization parameters, and outputting a pure noise image by the main noise reduction network; and carrying out difference processing on the pure noise image and the to-be-denoised image to obtain a denoised image.

Description

technical field [0001] The present disclosure relates to the technical field of image processing, in particular to an image denoising method, system, device and medium based on transfer learning. Background technique [0002] The statements in this section merely mention background art related to the present disclosure and do not necessarily constitute prior art. [0003] In the process of realizing the present disclosure, the inventors found that the following technical problems existed in the prior art: [0004] At present, with the development of society and the advancement of science and technology, electronic products such as computers, ipads, and mobile phones are becoming more and more popular. People expect to obtain better perceptual information from high-definition images. Bad lighting or high temperatures can cause sensor noise, resulting in images rendered with unnecessary noise. These noisy images need to be denoised, and the restored high-definition images wi...

AI Technical Summary

Problems solved by technology

[0008] For the training phase, the existing image denoising network uses the BN (Batch Normalization) layer to speed up convergence and control overfitting, allowing a larger learning rate to be used. BN calculates the mean and variance based on mini batch data, but BN There is a dependence on batchsize, when the batch_size is larger, the effect will be obvious

Existing image denoising networks use noise estimation as special guidance information, but do not take into account the influence of noise distribution information on network denoising

Images