WINNet: Wavelet-inspired Invertible Network for Image Denoising

Abstract

Image denoising aims to restore a clean image from an observed noisy image. The model-based image denoising approaches can achieve good generalization ability over different noise levels and are with high interpretability. Learning-based approaches are able to achieve better results, but usually with weaker generalization ability and interpretability. In this paper, we propose a wavelet-inspired invertible network (WINNet) to combine the merits of the wavelet-based approaches and learningbased approaches. The proposed WINNet consists of K-scale of lifting inspired invertible neural networks (LINNs) and sparsitydriven denoising networks together with a noise estimation network. The network architecture of LINNs is inspired by the lifting scheme in wavelets. LINNs are used to learn a non-linear redundant transform with perfect reconstruction property to facilitate noise removal. The denoising network implements a sparse coding process for denoising. The noise estimation network estimates the noise level from the input image which will be used to adaptively adjust the soft-thresholds in LINNs. The forward transform of LINNs produce a redundant multi-scale representation for denoising. The denoised image is reconstructed using the inverse transform of LINNs with the denoised detail channels and the original coarse channel. The simulation results show that the proposed WINNet method is highly interpretable and has strong generalization ability to unseen noise levels. It also achieves competitive results in the non-blind/blind image denoising and in image deblurring.

Reversible Vision Transformers

“…An interesting line of work builds upon the Reversible ResNets ideas proposing better reversible CNN models using ODE characterizations [6,39,59], momentum [39,59], layer-wise inversion [25], fourier transform based inversion [20] and fixed point iteration based inversion [2,60]. Reversible CNNs have been applied to several traditional image tasks such as compression [46], reconstruction [43], retrieval [42], and denoising [33,47] as well as to compressed sensing [61], compact resolution [75], image to image translation [67], remote sensing [56], medical image segmentation [55,74] and MRI reconstruction [57]. Reversible transformation have also been adapted to other networks such as RNNs [51], Unet [4,16], Masked Convolutional Networks [60] and 1000-layer deep Graph Neural Networks [40].…”

DURRNet: Deep Unfolded Single Image Reflection Removal Network

“…However, the fixed transform may not be sufficiently flexible to handle complex reflections. Inspired by the invertible networks as a learnable invertible transform [8,9], we propose to use the invertible networks to construct a learnable proximal operator ProxInvNete θ (•) for imposing the exclusion condition. The forward pass of the invertible networks serves as the forward transform, and the backward pass of the invertible networks then serves as the corresponding inverse transform.…”

“…In the Threshold Network (ThreNet), the feature of R will be concatenated with that of φ( T(k) ) and then they pass through a convolutional network with residual blocks to generate corrections for the feature of T. The updated feature of T will then be converted back to image domain using the backward pass of the invertible networks. Similar operations can be performed when updating R. The forward and backward pass of the invertible networks are constructed by the same set of P pairs of prediction and updater networks (PUNet), for details please refer to [9].…”