Underwater imaging is critically important for diverse scientific and engineering disciplines, including marine ecology, oceanography, subsea archaeology, and autonomous underwater vehicle (AUV) guidance. However, aquatic environments impose severe optical degradations on captured imagery through wavelength-selective light absorption, forward and backward scattering, non-uniform illumination, and suspended particulate interference. These phenomena collectively produce images characterised by chromatic distortion, attenuated contrast, and compromised structural definition. This paper proposes a Preprocessed Multiscale Fusion (PMF) framework that systematically addresses the deficiencies of prior single-image fusion approaches. The PMF method integrates physics-informed preprocessing stages—encompassing Shades-of-Gray white balancing, exponential wavelength compensation, homomorphic illumination correction, guided edge-preserving denoising, and Contrast-Limited Adaptive Histogram Equalization (CLAHE)—prior to multiscale fusion. Four complementary image representations derived from this pipeline are fused via transmission-guided adaptive weights through a Laplacian pyramid architecture. Comprehensive evaluation on the UIEB and EUVP benchmark datasets using both no-reference metrics (UIQM, UCIQE) and full-reference metrics (SSIM, PSNR) demonstrates that PMF achieves mean UIQM of 0.1936 (compared to 0.0808 for the baseline Ancuti et al. approach), representing a 139.6% improvement, and mean UCIQE of 38.10 (versus 28.53 for baseline), a 33.5% improvement. An ablation study confirms that each preprocessing component contributes incrementally to the final enhancement. The framework proves suitable for offline processing in marine robotics, ecological monitoring, underwater archaeology, and scientific documentation applications.
Kumbhar et al. (Thu,) studied this question.