CWFAS-Net: Low-light image enhancement using curvelet wavelet Attention and Fourier Transform

Abstract

Significant advancements have been made in low-light image enhancement techniques, however, challenges remain regarding inconsistent restoration quality and unsatisfactory visual perception. To address these issues, we propose a robust and efficient method, CWFAS-Net. First, an Amplitude Illumination Estimation Module (AIEM) is constructed to enhance global brightness by amplifying amplitude components in the frequency domain. Second, the Curvelet-Wavelet Fourier Attention (CWFA) and Detail-Enhancement Attention (DEMA) modules are designed. CWFA combines features from the Fourier and wavelet domains to improve texture detail recovery and overall visual quality, while DEMA extracts local spatial features using a detail-enhancement attention mechanism. Finally, an SNR map is incorporated as a prior to guide information fusion within CWFA and DEMA. Using eight public benchmark datasets, both reference and non-reference metrics demonstrate that CWFAS-Net surpasses most mainstream algorithms, delivering superior enhancement and generalization in low-light restoration, thus validating our approach.