Image Denoising Based On Shearlet Transform And Deep CNN

With the widespread use of image technology,image de-noising technology has penetrated into all aspects of human life.In the future,AI will become more popular and the lens will gradually deepen into everyday life.The demand for image clarity will also increase.Therefore,image de-noising has certain research value and broad prospects.In the process of image de-noising,how to remove noise effectively while protecting the details of the edge,and less time-consuming,has become a difficult problem to be solved.Therefore,this paper proposes two de-noising algorithms.One is Shearlet transform combined with popular compressed sensing de-noising algorithm.It not only obtains a good de-noising effect,but also has a high objective evaluation value.Another method utilizes the concept of deep learning.The training model can remove a certain range of noise images excellently and reduce training cost,which can solve the above problems better,so as to become the focus of the study of image de-noising algorithm.In this paper,we first introduce the typical image de-noising algorithms,and then analyze them.We complete the experiment simulation finally,and give the subjective and objective evaluation and the detail comparison.As following:1.In this paper,we researched and analysed the achievements of the predecessors.In order to solve the pseudo-Gibbs oscillation and fuzzy phenomena in the existing image de-noising algorithms,a new image de-noising algorithm is proposed by using the theory of image sparse representation in Shearlet domain.The proposed algorithm is to de-noise image from the entire image information:Based on the unique advantages of the Shearlet transfor,then the sparse representation model is used to construct the de-noising optimization model.The performed sparse coding,Monte_Carlo to select the clean Shearlet coefficients and finanlly reconstructed image.The experiment results show that the proposed method not only increase 3dB in PSNR,but also increase 0.03 in MSSIM,and makes the visual effect better,guarantees texture details and edge information.2.For the existing de-noising models,only the noise images participating in training can be removed,not remove the noisy images which not participate in training,and a certain range of image de-noising can be achieved.In this paper,an adaptive image de-noising based on convolutional neural network is proposed.No longer stick to the traditional Gaussian de-noising model,use the residual de-noising model to remove hidden noise when training the image,and use the threshold as an activation function to adjust the network model,and use BN and other algorithms to reduce the training time and improve the Noise performance.The experimental results show that the improved algorithm has good de-noising effect and good detail protection ability,and achieves image de-noising in a certain range and reduces the training cost.3.Matlab software is used to complete the experimental simulation,the GPU and Matconvnet learning framework are needed to support the proposed convolutional network de-noising model,and the effectiveness and superiority of the proposed algorithm are verified.The final simulation results are in accordance with the theoretical analysis conclusions,which shows that the proposed algorithm is superior to the traditional algorithm in overall performance,and has a wide range of application prospects and practical value.In the end,all the work of this paper is summarized and the future work of this field is also arranged.