数据同化框架下红外与可见光图像的可控融合  

Controllable fusion for infrared and visible light images within data assimilation framework

作　　者：陈荣元[1,2] 谢伟[2] 王四春[1] 秦前清[2] 

机构地区：[1]湖南商学院信息学院,管理工程研究所,湖南长沙410205 [2]武汉大学测绘遥感信息工程国家重点实验室,湖北武汉430079

出　　处：《光电子．激光》2010年第4期610-613,共4页Journal of Optoelectronics·Laser

基　　金：国家"973"计划资助项目(2006CB701303);国家"863"计划资助项目(2006AA12Z132);湖南省教育厅资助科研项目(08C485;09C567);应用型本科院校"十一五"国家课题资助项目(FIB070335-A8-17);湖南省科技厅资助项目(2009GK3131);湖南省社科基金重点资助项目(09ZDB080);湖南省自然科学基金资助项目(05JJ40103)

摘　　要：针对现有图像融合方法的规则不易根据融合图像的后续使用要求进行自适应调整,并且各种方法的优点不易综合的问题,提出一种数据同化框架下的基于粒子群优化(PSD)和差分进化(DE)算法的融合方法。该方法把比率低通金字塔变换作为数据同化系统的模型算子,把非下采样的Contourlet变换(NSCT,nonsubsampled contourlet transform)作为观测算子,根据后续处理对图像各个属性指标值的依赖程度构造目标函数,再利用PSD和DE的混合算法来优化目标函数从而获取合适的影像。两组实验从视觉效果和定量指标两方面验证了方法的有效性。Since existing fusion rules can not be adjusted adaptively according to the requirements of fol- low-up image processing steps and the merits of different fusion methods are not easy to be integrated, a controllable fusion for infrared and visible light images within data assimilation framework is proposed. Under this framework, a ratio pyramid transform is used as the model operator and a nonsubsampled contourlet transform as the observer operator first. Then the weights of different indices are determined according to their importance degrees to the following process and the objective function is composed of the weight sum of indices. Finally a particle swarm optimization and a differential evolution algorithm are employed to obtain the proper output image. Two groups of experiments have verified the validity of the proposed method in terms of both visual quality and quantitative indices.

关 键 词：图像融合 数据同化 粒子群优化(PSD)算法 差分进化(DE)算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]