Underwater image enhancement with zero-point symmetry prior and reciprocal mapping

Abstract

Images captured underwater typically exhibit color distortion, low brightness, and pseudo-haze due to light absorption and scattering. These degradations limit underwater image display and analysis, and still challenge the performance of current methods. To overcome these drawbacks, we propose a targeted and systematic method. Specifically, based on a key observation and extensive statistical analysis, we develop a Zero-Point Symmetry Prior (ZPSP): the histograms of channels a and b in the Lab color space, for color-balanced images, exhibit a symmetry distribution around the zero-point. Guided by the ZPSP, a Color Histogram Symmetry (CHS) method is proposed to balance color differences between channels a and b by ensuring they adhere to ZPSP. For channel L, a Reciprocal Mapping (RM) method is proposed to remove pseudo-haze and improve brightness, by aligning its reflectance and illumination components with the Dark Channel Prior (DCP) and Bright Channel Prior (BCP), respectively. Relatedly, it employs a divide-and-conquer strategy, distinguishing underwater image degradations in decomposed sub-images and tackling them individually. Notably, the above-proposed methods are integrated into a systematic enhancement framework, while focusing on targeted optimization for each type of degradation. Benefiting from the proposed strategy and methods, various degradations are individually optimized and mutually promoted, consistently producing visually pleasing results. Comprehensive experiments demonstrate that the proposed method exhibits remarkable performance on various underwater image datasets and applications, also showing good generalization ability. The code is available at: https://github.com/CN-lifei/ZSRM.