On Impulsive Noise Removal Algorithms Based On Deep Learning And Variational Model

Due to hardware reasons such as pixel failures in camera sensors and memory location failures in hardware,images are often contaminated with pulsed salt and pepper noise.Removing salt and pepper noise is an important problem in image processing.Firstly,we propose a new variational model based on tight frame for remov-ing salt and pepper noise.Our variational model uses a tight frame regulariza-tion term and an l₁-l₂non-convex fidelity term.By adding several auxiliary variables,we transform the original unconstrained problem into a constrained problem.Then we use the alternating direction multiplier method?ADMM?to get an efficient algorithm for solving the proposed model.In the experiment,we conduct a comparative analysis of the parameters of the model.The comparison with similar algorithms shows that our algorithm can achieve better denoising performance.Secondly,we propose a deep learning based method for removing salt and pepper noise.Based on the denoising convolutional neural network?DnCNN?,we introduce a new loss function for salt and pepper noise by considering the prior information of the noise mask.Because the mask information is taken into account during the training process,the learning process of the network becomes more efficient and the denoising performance is thus improved.The experimental comparison results on the three standard test datasets,our proposed method is much better than the traditional methods.Moreover,the proposed method outperforms DnCNN.