An Intelligent Underwater Image Enhancement Framework Based on Multi-Feature Fusion, Physical Models and Adaptive Restoration

Underwater image enhancement plays a pivotal role in advancing marine exploration and visual perception systems. To address the issue of deteriorating underwater image quality caused by light absorption and scattering during propagation in water, this study proposes a novel hybrid enhancement framework addressing three critical challenges in underwater imagery: color distortion, low illumination, and blur degradation. Our methodology integrates three innovative components: 1) A dual-space feature characterization system combining HSV color space analysis with Tenengrad-based blur detection, enhanced by Gaussian Mixture Model clustering and LVW-optimized feature selection for precise degradation pattern recognition. 2) A physics-guided image degradation model extending the J-M framework through adaptive RGB channel modulation and convolutional blur simulation, enabling realistic underwater condition emulation. 3) A two-stage enhancement architecture synergizing pixel-level color correction (via adaptive RGB coefficient mapping) with CLAHE-based illumination recovery and Multi-Scale Retinex with Color Restore(MSRCR), further augmented by Automatic Color Equalization(ACE) for visual consistency preservation. We use the datasets officially provided by the organizing committee of the Asia-Pacific Mathematical Contest in Modeling for verification. Experimental validation demonstrates significant performance improvements across both synthetic and real-world datasets. Quantitative evaluation using PSNR, UCIQE, and UIQM metrics reveals a 32.6% enhancement in color restoration accuracy compared to conventional methods. The proposed framework achieves superior edge preservation (15.8% higher Tenengrad scores) while maintaining natural color consistency through our adaptive channel balancing mechanism. Comparative ANOVA analysis confirms the statistical significance (p<0.01) of our method's superiority in handling complex underwater conditions. This study presents an interpretable framework integrating physical models with data-driven methods, applicable to underwater robotics and underwater communication systems. Our systematic sensitivity analysis informs future research directions for adaptive turbidity enhancement.