An Underwater Polarization Image Fusion Algorithm Based on Information Entropy and a Hierarchical-Adaptive Fusion Framework.

Underwater images often exhibit low contrast and loss of detail due to light scattering and absorption, which poses significant challenges for visual analysis in aquatic environments. Polarization imaging addresses these issues by exploiting the polarization states of light, effectively reducing backscatter and enhancing image contrast. In this paper, we propose a polarization image fusion method guided by information entropy and a hierarchical-adaptive fusion strategy. Local information entropy is first employed to perform multiscale denoising on Degree of Linear Polarization (DOLP) images, enabling adaptive detail reconstruction while distinguishing texture from noise. Subsequently, a hierarchical fusion framework is applied: low-frequency components are enhanced through detail injection, while high-frequency components are fused using a structure-guided mechanism that leverages low-frequency gradient information to generate soft masks for phase-aligned detail integration and edge sharpening. Experiments conducted on self-collected underwater images, two public underwater datasets, and three general-scene datasets demonstrate that the proposed method improves objective metrics, including information entropy, average gradient, and edge strength. Subjective evaluations further confirm its effectiveness in preserving details and adapting to diverse scenes. Furthermore, rigorous ablation studies and runtime analyses demonstrate that the optimized framework achieves a highly favorable balance between robust, artifact-free detail enhancement and computational efficiency. The proposed approach provides a practical solution for underwater image enhancement, with potential applications in target detection and infrastructure inspection.