Perceptual adversarial non-residual learning for blind image denoising

Remote Sensing Image Reconstruction Method Based on Parameter Adaptive Dual-Channel Pulse-Coupled Neural Network to Optimize Multiscale Decomposition

“…There are numerous methods for image denoising based on different theories. The reference [3] proposed Block Matching and 3D Filtering (BM3D) image denoising based on Non-local Self Similarity (NSS),, which performs similar block matching group denoising of images by minimum Euclidean distance; the reference [4] used Nuclear Norm Minimization (NNM) instead of low-rank matrix, and image noise reduction by finding the global optimal solution of non-convex optimized low-rank matrix, which has the problem of consistent treatment of singular values; the reference [5] proposed Weighted Nuclear Norm Minimization (WNNM) to choose different weights for singular values, which better recovers the low-rank matrix and has better denoising with better results; in the reference [6], a Denoising Convolutional Neural Network (DnCNN) was proposed, and the network learning object is the image residual; in the reference [7], the Fast and Flexible Denoising Convolutional Neural Network (FFDNet) was proposed on DnCNN basis, and the adaptability to different noises after adding noise levels is higher than that of DnCNN; The reference [8] proposed Wasserstein Generative Adversarial Networks (WGAN) and applied it to cell image denoising to solve the problem of blurred image texture details after CNN denoising, and achieved certain results; the reference [9] used generative adversarial networks and non-residual learning process for blind image denoising application to solve image artifacts and blurring problems; the reference [10] introduced an iterative correction scheme and proposed an effective guided feature domain denoising residual network for real-world noise denoising with some progress. Each theoretical method has different noise reduction effects, but basically, there are problems such as loss of image texture details, blurred edge contour structure, and insufficient image energy.…”