Abstract:Deep neural networks (DNNs) have achieved significant success in image restoration tasks by directly learning a powerful non-linear mapping from corrupted images to their latent clean ones. However, there still exist two major limitations for these deep learning (DL)-based methods. Firstly, the noises contained in real corrupted images are very complex, usually neglected and largely under-estimated in most current methods. Secondly, existing DL methods are mostly trained on one pre-assumed degradation process for all of the training image pairs, such as the widely used bicubic downsampling assumption in the image super-resolution task, inevitably leading to poor generalization performance when the true degradation does not match with such assumed one. To address these issues, we propose a unified generative model for the image restoration, which elaborately configures the degradation process from the latent clean image to the observed corrupted one. Specifically, different from most of current methods, the pixel-wisely non-i.i.d. Gaussian distribution, being with more flexibility, is adopted in our method to fit the complex real noises. Furthermore, the method is built on the general image degradation process, making it capable of adapting diverse degradations under one single model. Besides, we design a variational inference algorithm to learn all parameters involved in the proposed model with explicit form of objective loss. Specifically, beyond traditional variational methodology, two DNNs are employed to parameterize the posteriori distributions, one to infer the distribution of the latent clean image, and another to infer the distribution of the image noise. Extensive experiments demonstrate the superiority of the proposed method on three classical image restoration tasks, including image denoising, image super-resolution and JPEG image deblocking.

Networks for Modeling and Signal-Processing

Theoretical Foundation for Nonlinear Edge-Preserving Regularized Learning Image Restoration

Robust H∞ Filtering for Networked Stochastic Systems with Randomly Occurring Sensor Nonlinearities and Packet Dropouts

H∞Filtering over Networks for a Class of Discrete-Time Stochastic System with Randomly Occurred Sensor Nonlinearity

Hopfield Neural Network-Based Image Restoration with Adaptive Mixed-Norm Regularization

Hopfield Neural Network Based Color Image Restoration

Distributed Robust Bayesian Filtering for State Estimation.

Networked Strong Tracking Filtering with Multiple Packet Dropouts: Algorithms and Applications

Relaxing Tight Frame Condition in Parallel Proximal Methods for Signal Restoration

Hyperspectral Image Restoration under Complex Multi-Band Noises

Distributed <inline-formula> <tex-math notation="LaTeX">$H_\infty$ </tex-math> </inline-formula> State Estimation for a Class of Filtering Networks With Time-Varying Switching Topologies and Packet Losses

Hyperspectral Image Restoration Combining Intrinsic Image Characterization With Robust Noise Modeling

Neural Network based Linear and Nonlinear Noise Estimation

Variational Image Restoration Network.

Functional Neural Networks for Parametric Image Restoration Problems

An efficient and robust varying-parameter projection neural network for sparse signal reconstruction

Study Of Neural Networks In Image Restoration

Satellite Image Restoration via an Adaptive QWNNM Model

On a nonlinear nonlocal reaction-diffusion system applied to image restoration

Variational Image Restoration Network

A Robust PCA Approach with Noise Structure Learning and Spatial–Spectral Low-Rank Modeling for Hyperspectral Image Restoration