Abstract:As a fundamental task in imaging and computer vision, image denoising has remained an active topic of discussion over recent decades. Traditionally, a denoiser is designed by inferring analytical priors in the image,1,2 whose accuracy is restrained by its specified form. Recently, the rapid growth in computational hardware has reignited research on neural networks. Image denoising has also been widely investigated in relation to neural networks. Given sufficient image samples and advanced training techniques,3,4 convolutional neural networks (CNNs)5,6 have achieved more promising results than prior-based methods. However, current CNN denoisers remain far from achieving a realistic noise-free performance. First, real-image noise consists of complex noise components;7 therefore, it is highly Deep convolutional neural networks (CNNs) have achieved considerable success in terms of image denoising. However, previous CNN denoisers have been restricted by rigid kernel convolution that applies equal spatial treatment across images. To fully utilize the local differences, we propose a kernel prediction network that examines each pixel region and predicts unique pixel-wise kernels. Several optimizations have been further designed to gather sufficient information for single-image denoising task. We adopt dilated residual blocks to view the local pixel region at varying receptive fields. Then, kernel fusion assembles the information from different scopes and generates accurate kernels for each pixel. Instead of applying the predicted kernels to the original image, we construct a compressed feature map as a substitution such that more relevant local features are collected. Experiments are used to demonstrate that our network achieves favorable results compared with state-of-the-art methods and is adequate for practical applications. (c) 2021 SPIE and IS&T

Convolutional Neural Networks Analyzed via Inverse Problem Theory and Sparse Representations

Dilated kernel prediction network for

Deep Convolutional Neural Network for Inverse Problems in Imaging

Do CNNs solve the CT inverse problem?

Untrained Neural Network Priors for Inverse Imaging Problems: A Survey

Convergence Analysis for Deep Sparse Coding via Convolutional Neural Networks

On the unreasonable effectiveness of CNNs

Adaptive Loss Function for Super Resolution Neural Networks Using Convex Optimization Techniques

Convergence and Recovery Guarantees of Unsupervised Neural Networks for Inverse Problems

Theoretical Perspectives on Deep Learning Methods in Inverse Problems

The Neural Tangent Link Between CNN Denoisers and Non-Local Filters

To be or not to be stable, that is the question: understanding neural networks for inverse problems

NESTANets: stable, accurate and efficient neural networks for analysis-sparse inverse problems

Noise2Inverse: Self-supervised deep convolutional denoising for tomography

A Sparse Coding Interpretation of Neural Networks and Theoretical Implications

Denoising and Regularization via Exploiting the Structural Bias of Convolutional Generators

Deep neural networks can stably solve high-dimensional, noisy, non-linear inverse problems

Deeper Interpretability of Deep Networks

Invertible ResNets for Inverse Imaging Problems: Competitive Performance with Provable Regularization Properties

Randomness in Deconvolutional Networks for Visual Representation