文献类型：期刊文献

英文题名：Retinex-inspired underwater image enhancement with information entropy smoothing and non-uniform illumination priors

作者：Ma, Haiping[1];Huang, Jiyuan[1];Shen, Chenxu[2];Jiang, Zheheng[3]

机构：[1]Shaoxing Univ, Dept Elect Engn, Shaoxing 312000, Zhejiang, Peoples R China;[2]Hangzhou City Univ, Waikato Joint Inst, Hangzhou 310015, Zhejiang, Peoples R China;[3]Univ Leicester, Sch Comp & Math Sci, Leicester LE1 7RH, England

年份：2025

卷号：162

外文期刊名：PATTERN RECOGNITION

收录：SCI-EXPANDED(收录号：WOS:001414832300001)、、EI(收录号：20250517774697)、Scopus(收录号：2-s2.0-85216112273)、WOS

基金：This article was supported in part by the National Natural Science Foundation of China under Grant No. 61640316, and the Zhejiang Provincial Natural Science Foundation of China under Grant No. LY19F030011.

语种：英文

外文关键词：Underwater image enhancement; Retinex variational model; Non-uniform illumination; Minimum weighted error entropy; Independent and piecewise identical; distribution

外文摘要：Underwater image enhancement has attracted much attention in underwater vision. Recently, the enhancement methods based on Retinex variational models show remarkable performances for estimating reflectance and illumination of images. However, for underwater image enhancement with non-uniform illumination, it is still a challenging issue for the existing Retinex variational models. To overcome this limitation, this paper proposes a new Retinex variational model with information entropy smoothing and non-uniform illumination priors to enhance underwater images. First, underwater illumination is essentially structural and uneven, and it is modeled as an independent and piecewise identical distribution, which is a generic model to describe the complicated underwater illumination environment and accommodates the traditional Gaussian distribution as a special case. Second, assisted by the proposed illumination distribution, a minimum weighted error entropy criterion, which is an information-theoretic learning method, is introduced into the Bayesian Retinex model for accurately piecewise approximating the non-uniform illumination. Further, combined with the reflectance with the l1-norm describing fine-scale details and significant edges, the underwater image enhancement is turned into two separated subproblems, and their solutions can be simultaneously derived by an efficient half-quadratic optimization algorithm. In addition, a benchmark dataset is collected, including 780 real underwater images with different non-uniform illumination. Extensive experiments show that the proposed method obtains superior performance in terms of different quantitative metrics, and exhibits better visual quality than several existing state-of-the-art approaches.