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

Researchers have introduced two new methods for image denoising: VSANet, a view-aware sparse attention network, and Mixed-norm TV (MixTV), a technique that reduces noise by minimizing total variation.

VSANet is designed for light field (LF) image denoising, representing the 4D LF feature map as a unified spatial-angular token space and performing cross-view aggregation via locality-sensitive hashing-based sparse attention^[1]. This enables global feature interactions with linear complexity, effectively exploiting LF correlations across views and spatial locations. Meanwhile, MixTV is proposed as an effective method for removing various types of noise, with numerical experiments showing improved effectiveness and denoising quality compared to other total variation (TV) models^[2]. The TV method, which aims to reduce noise by minimizing the total variation of the image, has been widely applied in image processing and computer vision for its ability to preserve edges and enhance image quality. Experiments with VSANet demonstrate that it outperforms state-of-the-art LF denoising methods. The MixTV model admits a unique solution, and the associated numerical algorithm guarantees convergence.