Hierarchical Wavelet Decomposition Network for Water-Related Optical Image Enhancement

Enhancing water-related optical images poses a significant challenge due to the complex interplay of direct attenuation and backscattering. Current methods primarily focus on modifying the spatial domain and pay less attention to the heterogeneity of the frequency domain degradation distributions, which limits their effectiveness in solving multiple types of degradation problems simultaneously. To overcome these limitations, we propose a hierarchical wavelet decomposition network (HWD-Net). HWD-Net leverages wavelet transforms to create a compact feature space, enabling the distinct restoration of low and high-frequency degradations through a strategic divide-and-conquer approach, which prevents the interaction of high- and low-frequency information and avoids the generation of incorrect textures. Furthermore, HWD-Net employs a hierarchical decomposition paradigm to progressively extract richer high-frequency information, achieving superior enhancements in a coarse-to-fine manner. Comprehensive evaluations on multiple underwater data sets demonstrate the superiority of HWD-Net over state-of-the-art methods in terms of image quality and inference time.