GLRF: Underwater Image Enhancement via Global and Local Visual Reconstruction Fusion

Abstract

Underwater images frequently suffer from severe quality degradation caused by complex environmental conditions and the scattering and absorption of light during transmission, which greatly restricts their utility in subsequent analysis and applications. To tackle this challenge, this article presents an underwater image enhancement method that converges global and local visual reconstruction fusion to improve image quality and accuracy of visual tasks, called GLRF. Specifically, we first implement global color correction using the attenuated channel compensation strategy and correct local color bias using the local overlap-weighted color compensation strategy. Subsequently, we propose a structure-constrained luminance reconstruction method with smooth brightness transitions and gradient constraints for global contrast enhancement, which reduces noise caused by uneven brightness and preserves image details, while traditional histogram-based methods may introduce noise and cause irreversible detail loss when adjusting the brightness distribution. Ultimately, we exploit wavelet decomposition and reconstruction to unsharp mask the low-frequency components and adaptively threshold denoise the high-frequency components to reconstruct a high-visibility result. The comprehensive experiments on four benchmark datasets have shown that our GLRF achieves competitive or superior performance on most evaluated metrics. Additionally, the results of our GLRF enhancement are beneficial for feature matching.