多区域融合的全卷积颜色恒常算法  

Full Convolution Color Constancy Algorithm for Multi-region Fusion

作　　者：刘凯 孙鹏[1,2] 倪新龙 赵瑞敏 解梦达 LIU Kai;SUN Peng;NI Xinlong;ZHAO Ruimin;XIE Mengda(Department of Criminal Science and Technique,Criminal Investigation Police University of China,Shenyang 110854,China;Key Lab of Forensic Science,Ministry of Justice,China(Academy of Forensic Science),Shanghai 200063,China;College of Forensic Sciences,Criminal Investigation Police University of China,Shenyang 110854,China;School of Cyberspace Security,Guangzhou University,Guangzhou 510006,China)

机构地区：[1]中国刑事警察学院公安信息技术与情报学院,沈阳110854 [2]司法部司法鉴定重点实验室,上海200063 [3]中国刑事警察学院刑事科学技术学院,沈阳110854 [4]广州大学网络空间安全学院,广州510006

出　　处：《小型微型计算机系统》2025年第3期655-661,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61307016)资助;辽宁省研究生教育教学改革研究项目(LNYJG2023317)资助;公安部技术研究计划项目(2020JSYJC25)资助;司法部司法鉴定重点实验室开放课题项目(KF202317)资助;“新时代犯罪治理研究中心”智库项目(2022007)资助;中国刑事警察学院研究生创新能力提升项目(2022YCYB50)资助.

摘　　要：针对现有颜色恒常性算法在复杂场景下的适用性差,光源估计精度不高的问题,提出一种多区域融合的全卷积颜色恒常算法.首先,利用金字塔场景解析网络(Pyramid Scene Parsing Network,PSPNet)模型对目标图像进行场景语义分割;其次,将本文提出的基于全卷积网络架构的端到端光源估计模型(Full Convolutional Illuminant Estimation Network,FCIEN)用于各语义区域进行光源估计,各区域通过加权融合得到整体图像的光源估计值;最后,根据得到的光源估计值使用Von Kries模型对目标图像进行偏色校正.在Gehler-Shi、NUS-8camera等数据集中的实验结果表明,相较于同类型的算法,本文算法的光源估计角度误差值分别下降20.73%、15.22%.实验结果表明,本文算法针对复杂场景的光源估计有着较强的鲁棒性和准确性.Aiming at the problems of poor applicability of existing color constancy algorithms in complex scenes and low accuracy of illuminant estimation,a full convolution color constancy algorithm for multi-region fusion is proposed.Firstly,the Pyramid Scene Parsing Network(PSPNet)model was used to segment the scene semantics of the target image.Secondly,the Full Convolutional Illuminant Estimation Network(FCIEN),an end-to-end estimation model based on the convolutional network architecture proposed in this paper,is used to estimate the illuminant in each semantic region.Finally,according to the estimated light source,Von Kries model is used to correct the color deviation of the target image.Experimental results in Gehler-Shi,NUS-8camera and other data sets show that compared with other algorithms of the same type,the Angle error of illuminant estimation of the proposed algorithm decreases by 20.73%and 15.22%respectively.The experimental results show that the algorithm can effectively solve the problem of illumination variation in complex scenes.

关 键 词：图像处理 颜色恒常性 语义分割 多区域融合 光源估计 场景语义 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]