Image noise reduction method

Abstract

The invention discloses an image noise reduction method, which includes: constructing a paired training data set and a verification data set according to the acquired natural image and the additive noise sampled in a Gaussian distribution with known variance, and Preprocess the training data set; use the convolutional neural network as the Boosting unit to build a depth boosting frame model based on the SOS algorithm; use the preprocessed training data set to train the depth boosting frame model and adjust the corresponding model parameters; use The verification data set adjusts the structural hyperparameters and optimization hyperparameters of the trained deep lifting framework model; then uses the verification data set to verify the deep lifting framework model, and selects the model parameters with the smallest recovery loss to determine the final deep lifting framework model ; image denoising using the final depth boosting framework model. Using the above method can improve the performance of image noise reduction.

Description

technical field [0001] The invention relates to the technical field of digital image processing, in particular to an image noise reduction method. Background technique [0002] There are complex noises in the process of digital image acquisition, including thermal noise of electronic components, sensor reading conversion noise, signal transmission noise, etc. [0003] Usually, the independent and identically distributed zero-mean Gaussian model v~N(0,σ 2 ) is used to model this additive noise. Let the image signal be x, then the image polluted by noise can be expressed as y=x+v. To reduce noise in images, methods based on image prior models have been extensively studied. For example, nonlocal approximation models (A.Buades, B.Coll, and J.M.Morel, “Nonlocal image and moviedenoising.” in International Journal of ComputerVision 2008, pp.123-139), sparse representation models (M.Elad, and M .Aharon, "Image denoising viasparse and redundant representations over learned dictio...

AI Technical Summary

Problems solved by technology

However, there is still a large performance gap in the boosting algorithm based on the classical model, and its advantages have not been fully exploited

Images