基于颜色转移和自适应增益控制的混合水下图像增强  

Hybrid Underwater Image Enhancement Based on Color Transfer and Adaptive Gain Control

作　　者：崔峥 王森[1] 王娴[1,2] 段浩 杨春曦 那靖 CUI Zheng;WANG Sen;WANG Xian;DUAN Hao;YANG Chun-Xi;NA Jing(Faculty of Mechanical and Electrical Engineering,Kunming University of Science and Technology,Kunming 650500;Yunnan Key Laboratory of Intelligent Control and Application,Kunming 650500)

机构地区：[1]昆明理工大学机电工程学院,昆明650500 [2]云南省智能控制与应用重点实验室,昆明650500

出　　处：《自动化学报》2025年第2期376-390,共15页Acta Automatica Sinica

基　　金：国家重点研发计划(2023YFE0204700);国家自然科学基金(62433012,62063011);云南省科技厅重大专项项目(202302AD080005)资助。

摘　　要：针对水下图像的颜色偏差和低对比度等退化问题,提出一种基于颜色转移和自适应增益控制的混合水下图像增强方法.首先,根据颜色转移图像和最大衰减图引导的融合策略校正水下图像的颜色偏差.其次,利用一阶原始对偶方法对V通道进行滤波以有效地抑制噪声的干扰,获得结构图像;并且提出自适应增益控制,根据图像的高频信息自适应调整增益,以获得细节图像.最后,通过加权融合结构图像与细节图像,得到高质量的水下图像.实验结果表明,针对不同自然和工业环境下的水下图像:1)所提方法可以有效地校正颜色失真现象;2)显著提高水下图像的对比度并且抑制噪声干扰;3)在定量评价指标和高级视觉任务(目标检测、图像分割、关键点检测和水下双目测距)中优于其他主流水下图像增强方法,为水下目标抓取等工程应用提供了有益的参考.Aiming at handling the degradation problems of color deviation and low contrast in underwater images,a hybrid underwater image enhancement method based on color transfer and adaptive gain control is proposed.Firstly,the color transfer image and the maximum attenuation map-guided fusion strategy are used to correct the color deviation of underwater images.Secondly,the structure image of the V channel is obtained based on the firstorder primal-dual algorithm to effectively suppress the noise interference.Furthermore,adaptive gain control based on the high-frequency information of the image is proposed to adaptively adjust the gain to obtain detailed image.Finally,the structure image and the detail image are weighted fused to obtain a high-quality underwater image.The experimental results demonstrate that for underwater images with various natural and industrial environments,1)The outputs of the proposed method can effectively correct the color distortion;2)It can improve the contrast of underwater images and suppress the noise interference;3)The quantitative metrics and high-level visual tasks(object detection,image segmentation,key-point detection,and underwater binocular distance measurement)are superior to other state-of-the-art underwater enhancement methods,which provides a useful reference for engineering applications such as underwater automated grasping.

关 键 词：水下图像增强 颜色校正 梯度平滑 纹理增强 水下成像模型 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]