Research On Image Denoising Based On Joint Priors

In the past 40 years,the performance of existing image denoising methods is still not ideal To this end,through joint prior model,this thesis proposes two image denoising methods.One is based on structure-constrained low-rank approximation,and the other is based on detail-injection convolutional neural network.Through simulation the noisy images and collect real images,the experimental verification shows that the methods in this thesis can effectively restore the structural details of images while removing noise,which are usually superior to the most advanced image denoising methods(1)The image denoising method based on structure-constrained low-rank approximation as a model-driven method,it combines prior low-rank regularization term based on structural self-similarity,with introducing shape-aware kernel functions to design kernel wiener filtering regularization term and establish an image denoising model based on global and local priors Alternating direction multiplier method(ADMM)is used to decompose this complex non-convex optimization problem into three sub-problems,and then the noise-removed image is finally obtained by solving one by one and loop iteration.The experimental results show that the denoising effect using this method is better than the most advanced model-based image denoising method,and sometimes even exceeds the image denoising method based on deep learning(2)The image denoising method based on detail-injection convolutional neural network the abovementioned image denoising method that is based on structure-constrained low-rank approximation has modeling bottlenecks caused by manually characterizing features and is complicated to adjust the model parameter,which is especially difficult to apply to the denoising of images of large data information.In order to overcome the limitations of the model-based method for image denoising,this thesis uses wavelet transform to extract high-frequency information of image signals as a local prior,combined with Spatial convolutional neural network as a global prior,and designs a multi-scale learnable activation function to replace the traditional one,and proposes a detail-injection convolutional neural network model for image denoising.Through a lot of simulation and experimental verification,the results show that this method can not only effectively remove noise of images,but also restore fine image structure.Compared with the most advanced image denoising methods,it usually obtains better subjective and objective denoising effects.