VSANet: View-aware Sparse Attention Network for Light Field Image Denoising

2026-06-23

AI Summary

This work addresses the challenge of light field image denoising, where noise is independently distributed across high-dimensional data yet exhibits strong correlations among views. To tackle this, the authors propose VSANet, which introduces a novel View-aware Sparse Attention (VSA) mechanism that enables global cross-view interactions with linear computational complexity within a unified spatial-angular feature space. Additionally, a Multi-subspace Feature Refinement (FR) module is designed to jointly integrate spatial, angular, and epipolar plane information, thereby enhancing multidimensional feature representation. Extensive experiments demonstrate that the proposed method consistently outperforms current state-of-the-art approaches across multiple benchmarks, achieving significant improvements in light field denoising performance.

Abstract

Light field (LF) image denoising is challenging due to the high-dimensional structure of LF data. While noise is independent across sub-aperture images, scene content exhibits strong cross-view correlations. We introduce VSANet, a view-aware sparse attention network for LF denoising. Specifically, we propose a view-aware sparse attention (VSA) block that represents the 4D LF feature map as a unified spatial-angular token space and performs cross-view aggregation via locality-sensitive hashing-based sparse attention. This enables global feature interactions with linear complexity, effectively exploiting LF correlations across views and spatial locations. In addition, we design a feature refinement (FR) block to emphasize informative features in spatial, angular, and epipolar subspaces. The VSA and FR blocks are integrated within a sequential attention refinement module, forming the core of VSANet. Experiments demonstrate VSANet outperforms stateof-the-art LF denoising methods.

Problem

• light field denoising
• high-dimensional data
• cross-view correlation
• sub-aperture images
• image noise

Innovation

• view-aware sparse attention
• light field denoising
• locality-sensitive hashing
• spatial-angular token space
• feature refinement