Lightweight Large Receptive Field Network via Wavelet Convolution for Nighttime Flare Removal

The task of nighttime flare removal aims to restore images damaged by flares, making it a new and valuable image restoration task with significant practical applications. During the flare restoration process, flares can be divided into large-area smooth streaks and high-contrast glare and shimmer types. Different types of flares occupy a large portion of the image and severely distort the background details, this requires the network to have a sufficiently large receptive field to handle large flares and the ability to recover fine local details. While some progress has been made in the field of nighttime flare removal, there is still a challenge in balancing receptive field size, parameter count, and local detail recovery in different approaches based on transformers and convolutions. To address this, we propose a nighttime flare removal network based on wavelet convolution (WCNFRN). To achieve a lightweight network with a large receptive field, we introduce a Wavelet Convolution Attention Module (WCAM) in the network. This module improves flare removal performance by combining a symmetric network structure, wavelet convolution groups, and a hybrid attention mechanism.