Hierarchical Iterative Image Denoising Based On Wavelet And Sparse Preprocessing

When the image is formed and transmitted,it will inevitably be disturbed by the noise.The noise in the image is often interwoven with the signal,which will obscure the details of the image such as the boundary contour and the lines.Wavelet transform can achieve image noise reduction by setting wavelet coefficients,while sparse compression determines image noise reduction by setting sparse coefficients.In this paper,wavelet transform and sparse compression algorithm are combined to make independent component analysis and neural network CNN optimization.Using the independent component analysis algorithm,the noise image can be treated as a mixture of noise signals to separate the signal,but the noise reduction method requires multiple observation signals to run.In order to solve the problem that independent component analysis and calculation performed by a single observation signal,a method of sparsely generating a plurality of observation signals for redundant information of a single image is proposed.Firstly,to make the only one noisy image to be sparse using the dictionary compression algorithm of KSVD(Kernel Singular Value Decomposition)and secondly,get the first-time denoised image using the redundant information.Finally,to make both the first-time denoised image and original noisy image as the multiple observations for ICA separation.Another kind of denoising method which is improved by using the sparse compression technique is based on CNN denoising technology.During the process of noise image training,this paper learns from multiple source signal images and different intensities of noisy images to obtain one Noise signal data model which complete the image noise reduction.In addition,the more learning process and images will lead to slow learning process of neural network.In order to solve this problem,the input image can be fully utilized without reducing the detail information through the dictionary sparse compression without reducing the image information of the image sparse and filtering redundant information.