基于RETINEX的水下图像增强融合算法  

RETINEX-based Fusion Algorithm for Underwater Image Enhancement

作　　者：巩依典 诸云[1] 郭佳 王红超[2] GONG Yidian;ZHU Yun;GUO Jia;WANG Hongchao(School of Automation,Nanjing University of Science and Technology,Nanjing210094,China;Xiamen Key Laboratory of Intelligent Fishery,Xiamen Ocean Vocational College,Xiamen 361100,China)

机构地区：[1]南京理工大学自动化学院,南京210094 [2]厦门海洋职业技术学院智慧渔业重点实验室,厦门361100

出　　处：《无人系统技术》2024年第6期84-92,共9页Unmanned Systems Technology

基　　金：厦门市智慧渔业重点实验室2023年度开放课题(XMKLI-0P-202301);湖州市城市多维感知与智能计算重点实验室2024年度开放基金(UMPIC202401);智能机器人湖北省重点实验室开放基金(430073)。

摘　　要：针对自主式水下航行器目标探测效率低下的问题,提出了一种基于Retinex的水下图像增强融合算法,旨在改善水下成像中因光照强度因素引起的色彩偏差和图像模糊问题。首先,采用红暗通道先验理论实现图像去雾;其次,使用红通道滤波来解决水下红光不足的问题;然后,结合自适应多尺度视网膜增强与颜色恢复算法,利用拉普拉斯锐化和色彩恢复因子进行色彩校正;最后,通过实验和主客观分析验证所提算法的可行性和优越性。实验结果表明,所提算法相较于自适应多尺度视网膜增强与颜色恢复算法的水下彩色图像质量评估指标提高了35.56%,相较于原始图像信息熵增加了19.03%,明显改善了水下图像的高模糊度和色彩失衡问题,显示出良好的处理效果和实用性,同时结合YOLO框架证明了所提算法在水下图像目标识别领域的实用性。To address the inefficiency of target detection in Autonomous Underwater Vehicles,a Retinex-based underwater image enhancement fusion algorithm is proposed.The algorithm focuses on mitigating color deviation and image blurring caused by varying light intensity in underwater environments.Firstly,the red and dark channel a priori theories are utilized for image defogging.Next,red channel filtering addresses the issue of insufficient red light underwater.Adaptive multi-scale retinal enhancement is combined with color recovery algorithms,using Laplace sharpening and color recovery factors for effective color correction.The proposed algorithm is validated through experiments and comprehensive analyses,both subjective and objective.Results demonstrate a 35.56%improvement in the underwater color image quality assessment index compared to the adaptive multi-scale retina enhancement algorithm,and a 19.03%increase in information entropy compared to the original image.The method significantly reduces blurriness and corrects color imbalance in underwater images,proving its effectiveness and practicality.Furthermore,when integrated with the YOLO framework,the algorithm shows strong potential for underwater target recognition.

关 键 词：水下图像处理 红色暗通道先验理论 自主式水下航行器 RETINEX算法 色彩失衡 图像评价指标 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]