Research On Transformation Analysis Image Denoised Based On U-net Structure

With the development of computer vision technology,the image denoising technology has received more and more attention.Although the traditional denoising algorithm based on multiple transformation methods can remove noise to a certain extent,it cannot make good use of the advanced features in the image,resulting in great uncertainty in the final denoising effect.Therefore,in this paper,a denoising algorithm which combines the U-net structure with multiple transform methods has been proposed for Gaussian noise.The algorithm can make full use of image feature to improve image clarity and reduce computational costs.The specific research content include :(1)This paper proposes a U-net-based Feature Fusion Network,the network performs multi-scale feature extraction and fusion on the image at the feature extraction layer,and strengthens the ability to suppress noise based on extracted features.In the U-net structure,the down-sampling is replaced with a combination of wavelet transform and principal component analysis,and the up-sampling is replaced with inverse wavelet transform to improve the feature extraction ability of the network.In addition,an experimental analysis is performed on the existing activation function,and finally an activation function suitable for the network is determined.Experimental verification shows that the denoising network can make better use of image feature information and improve the clarity and detail recovery capabilities of noisy images.(2)Based on the further optimization of the above-mentioned denoising network,this paper proposes a U-net-based Dense Residual Network.In order to improve the convergence speed and training effect of the network,the normalization method and optimization algorithm adopted by the network are determined through experimental analysis.Dilated convolution is used to improve the receptive field of the network,so as to improve the final denoising effect of the network.It is verified by experiments that the denoising network can reduce the computational cost and enhance the denoising effect of the network while making full use of the image feature information.