Research On Image Denoising Algorithm Based On Hybrid Domain And Residual Network

In recent years,pictures have played an important role in information transmission in human life,and the size of picture noise directly affects people's viewing experience.Therefore,how to better denoise pictures has become an urgent problem to be solved.Image denoising is a long-standing problem in the image field.The solution methods developed to the present are mainly divided into traditional algorithms and deep learning-based methods.Most of the traditional algorithms are for simple noise,and the amount of calculation is relatively small,but there is also denoising.The effect is weak and some algorithm parameters are reduced.Through the increase of data volume and the enhancement of computer computing power,many algorithms based on neural networks have been proposed.Such algorithms have generally better denoising performance than traditional algorithms,but they are computationally intensive and timeconsuming.Aiming at the shortcomings of the above traditional algorithms and deep learning algorithms,this paper conducts research in two types of algorithms.In view of the simple denoising of traditional algorithms and the complicated adjustment of hyperparameters,this article first conducts denoising experiments for different algorithms in the space domain and frequency domain on the basis of noise classification analysis,and obtains better denoising effects of NLM and denoising in frequency domain The results are suitable for periodic noise,and then this article conducts hyperparameter adjustment experiments for the 2D and 3D thresholds in the BM3 D algorithm,and the PSNR is the best result of 29.735.In view of the large amount of calculation of the deep learning algorithm,this paper is based on the DNCNN algorithm and analyzes the reduction of the parameters of the original network structure by reducing the network depth and setting the grouping convolution.This paper also addresses the problem that some deep learning algorithms have general denoising effects and are difficult to train.By learning from previous experience,a novel model-based CNN denoiser is proposed,which improves denoising performance and increases ease of training by expanding convolution.At the end of this paper,denoising experiments are carried out on the three types of algorithms of BM3 D,DNCNN,and the proposed network.The experiment shows that the overall performance of the network in this paper is better than the BM3 D algorithm in terms of PSNR parameters.Some scenes perform better than the DNCNN algorithm.When the noise variance is 15,the PSNR of 26.198 db is reached,and the algorithm is very competitive,and it can be widely used in the picture denoising tasks of human daily life.Figure 54;Table 12;Reference 72.