Underwater image enhancement with frequency-spatial cross-domain transformer and hybrid collaborative representation

Underwater image enhancement is a pivotal technique for marine exploration and engineering. However, underwater images often suffer from severe degradation caused by wavelength-dependent attenuation, scattering, and refraction, which pose considerable challenges to existing methods. To address this issue, we propose a model that integrates a frequency-spatial cross-domain transformer and hybrid collaborative representation for underwater image enhancement. First, by introducing the frequency dimension, a cross-domain transformer equipped with an efficient feature interaction mechanism is designed. It simultaneously integrates spatial-domain and frequency-domain information to enrich feature representation and improve feature utilization efficiency via a fusion attention mechanism. Second, a hybrid frequency-spatial feature extraction block, which combines the cross-domain transformer with a feature attention block, is constructed as a dual-branch structure to concurrently capture global structures and local details. This enables a transition from single-domain feature modeling to cross-domain complementary feature representation. Third, a two-stage hybrid collaborative representation block is proposed. By facilitating refined interaction and fusion of global structures and local details, it achieves adaptive and dynamic feature collaboration, effectively suppressing noise, enhancing texture details, and correcting color casts. Experiments on five public datasets demonstrate that our approach outperforms state-of-the-art methods in both quantitative metrics and subjective visual perception, with an average relative PSNR improvement of 2.932%, highlighting its exceptional learning and generalization capabilities. Furthermore, its outstanding performance in advanced downstream tasks, including underwater target detection, salient object detection, and atmospherically degraded image enhancement, further validates its practical value and application potential in real-world marine engineering scenarios.