Enhancing Underwater Imagery through Multi-stream Pre- processing and Wavelet Decomposition

Underwater image acquisition is inherently affected by wavelength-dependent absorption and scattering, which reduce contrast, distort colors, and obscure fine details. These degradations hinder accurate interpretation in applications such as seabed exploration, underwater robotics, and ecological monitoring. To address these limitations, this paper presents a unified novel framework for enhancing color-degraded underwater images through a multi-stream pre-processing and wavelet-based fusion strategy. The pipeline integrates gamma correction with Gaussian filtering to suppress color imbalance and haze, histogram equalization to redistribute intensities and improve dynamic range, and median filtering to reduce impulsive noise while preserving edges. Outputs from these parallel streams are subsequently decomposed into wavelet subbands, followed by mean-based subband fusion and inverse reconstruction. Finally, a percentile-based RGB stretching step restores balanced brightness and dynamic contrast. Experimental evaluation was conducted on the Underwater Image Enhancement Benchmark (UIEB) dataset using both full-reference and non-reference quality metrics. Results show that the proposed method achieves superior visual quality and objective performance, with an average Peak Signal-to-Noise Ratio (PSNR) of 33.31 dB & Underwater Color Image Quality Evaluation (UCIQE) score of 0.798. The findings highlight the framework’s effectiveness in producing natural, high-clarity underwater imagery suitable for practical deployment.