Attenuation-aware hybrid learning framework for spectrally consistent underwater image enhancement

Underwater imagery is often degraded by severe color imbalance, low contrast, and detail loss caused by wavelength-dependent light absorption and scattering. To address these limitations, this work introduces a hybrid learning framework that integrates adaptive color compensation with layer-decomposed contrast optimization for achieving physically consistent and perceptually balanced restoration. The adaptive color compensation stage employs channel-wise spectral correction guided by the Beer–Lambert absorption model and a learning-based adaptive mapping to preserve chromatic accuracy across diverse lighting conditions. The layer-decomposed contrast optimization module further refines contrast through tone-scaled illumination and sparsity-regularized reflectance estimation while enhancing natural color tones and fine structural retention. Experimental validation on multiple underwater datasets demonstrates that the proposed method consistently achieves superior quantitative scores across entropy, UIQM, UCIQE, and PCQI metrics while maintaining the lowest BRISQUE index and reduced computational overhead. The proposed framework achieves an average processing time of 0.25–1.1 s per image, maintaining a strong balance between enhancement fidelity, perceptual realism, and computational efficiency for real-time underwater vision applications.