Research On Noise Level Estimation Method And Denoising Algorithm For Real Scene Image

In the process of image shooting,it is inevitable to be affected by camera equipment,acquisition environment,and transmission channel,resulting in obvious noise in the acquired image,which interferes with people’s understanding and analysis of image information.To ensure that the image content can be accurately expressed,it is necessary to denoise it after acquisition and transmission to reduce erroneous information on the image.Some recent studies have achieved remarkable results in removing additive white Gaussian noise,but it is still difficult to apply to real scene images.Because in real life,the noise distribution on the image is uncertain,and the algorithm for removing Gaussian white noise cannot remove the real noise very well.On the other hand,the depth of the denoising model is increasing,but the denoising effect cannot be significantly improved.For the characteristics of noise on real images,a simple but effective two-stage depth image denoising algorithm is designed.The algorithm estimates the noise distribution level in real images based on the attention mechanism and uses a multi-scale denoising module of hybrid dilated convolution kernel common convolution for non-blind denoising.In order to further improve the performance of the network,this paper improves the image denoising algorithm.In the stage of noise estimation,the global attention mechanism is used to extract multi-dimensional noise features.In the stage of non-blind denoising,a multi-branch structure is used to enhance the denoising ability of the network,and the reasoning time of the network is reduced by the reparameterization method.The algorithm was compared with the classical denoising algorithm on four datasets: SSID dataset,Nam dataset,Poly U dataset,and DND dataset.For comparison,peak signal-to-noise ratio(PSNR)and structural similarity(SSIM)were selected as the evaluation indicators of denoising effectiveness.The experimental results show that the two-stage real image denoising network designed in this paper improves both the evaluation index and the visual quality.At the same time,comparative experiments on SSID data sets were carried out to verify the effectiveness of the proposed algorithm.The comparative experiments show that the proposed algorithm can effectively improve the real image denoising effect by jumping connections,noise level estimation,and multi-scale modules.Compared with existing methods,the proposed algorithm can not only effectively remove real image noise,but also control the computational efficiency of the denoising network through simple module parameter settings.