Heterogeneous window transformer for image denoising

TL;DR

This paper introduces HWformer, a heterogeneous window transformer that efficiently captures global and local image information for denoising, achieving comparable results to existing methods with significantly reduced processing time.

Contribution

HWformer innovatively combines global window shifting and sparse local information extraction to enhance denoising performance efficiently.

Findings

HWformer reduces denoising time by 70% compared to Restormer.

Global window shifting improves context modeling without increasing computation.

Sparse local extraction prevents information loss across patches.

Abstract

Deep networks can usually depend on extracting more structural information to improve denoising results. However, they may ignore correlation between pixels from an image to pursue better denoising performance. Window transformer can use long- and short-distance modeling to interact pixels to address mentioned problem. To make a tradeoff between distance modeling and denoising time, we propose a heterogeneous window transformer (HWformer) for image denoising. HWformer first designs heterogeneous global windows to capture global context information for improving denoising effects. To build a bridge between long and short-distance modeling, global windows are horizontally and vertically shifted to facilitate diversified information without increasing denoising time. To prevent the information loss phenomenon of independent patches, sparse idea is guided a feed-forward network to extract local information of neighboring patches. The proposed HWformer only takes 30% of popular Restormer in terms of denoising time.