【作者】 孙欣；

【导师】 那彦；

【作者基本信息】 西安电子科技大学 ， 电路与系统， 2014， 硕士

【摘要】 本文主要研究基于粒子群优化的多传感器图像融合方法。文中首先介绍了粒子群算法的理论基础，给出了标准粒子群算法的工作流程以及基本步骤，接着对图像融合中用来分解源图像的多尺度几何分析工具进行了详细的理论分析。其次对图像融合的基本概念和融合图像的质量评价方法作了介绍。最后，在上述理论基础上，将“融合图像的质量评价”问题看作一个多目标优化问题，提出一种以多幅基于多尺度变换融合算法得到的融合图像组成初始粒子种群，以两个融合质量评价指标为寻优目标的基于改进的粒子群优化算法的图像融合方法。本文提出的融合方法在对不同类型的源图像进行性能优化时，根据不同的成像机理和融合目的，选取不同的评价指标作为多目标优化的目标函数。对于多聚焦图像，获得同一场景所有目标均聚焦清晰的图像是多聚焦图像融合技术的目标，这里选取平均梯度和信息熵作为目标函数；对于医学图像，为了获得清晰的骨骼组织和软组织显示效果，我们选择平均梯度和均方根误差作为目标函数；对于遥感图像，综合考虑光谱信息和细节信息的保留度，选取光谱相关系数和平均梯度作为目标函数。最后，基于文中提出的融合方法，对多传感器图像融合技术的三个重要领域—多聚焦图像、医学图像和遥感图像进行仿真实验，定性与定量结果一致说明文中方法具有更好的融合性能。更多还原

【Abstract】 The method of multi-sensor image fusion based on particle swarm optimization(PSO) is mainly discussed in the paper.Firstly, the theoretical basis of PSO algorithm is introduced, and the workflow andbasic steps of the standard PSO are given. Then Multi-scale Geometric Analysis (MGA)tools used for the decomposition of source images in image fusion are analyzed in detailin theory. Secondly, the basic concepts of image fusion and evaluating methods forquality of fused images are introduced. Finally, based on the theory above, taking thequality evaluation of the fused image problem as a multi-objective optimizationproblem, this paper proposes a method of image fusion based on a improved PSO, thismethod takes multiple fused images received from fusion algorithm based onmulti-scale transform as initial particle swarm and takes two evaluation indicator asobjectives for optimization.When different types of source images are processed for performance optimizationwith the fusion method the paper proposes, according to different imaging mechanismand fusion purposes, different evaluation indicators are selected as the objectivefunctions of the multi-objective optimization. In the case of multi-focus image, becausethe target of the multi-focus image fusion technology is to obtain a new image where alltargets in the same scene are in focus; in the case of medicine image, in order to getclear bone and soft tissue display, average gradient and root mean square error areselected as objective functions; in the case of remote sensing image, considering theretention degree of spectral information and detail information, spectral correlationcoefficient and average gradient are selected as the objective functions.Finally, based on the fusion method the paper proposed, experiments concerningthe three key areas of multi-sensor image fusion technology multi-focus image,medicine image and remote sensing image are done, Simulation results and quantitativeevaluating parameters consistently show the algorithm has better fusion performance.更多还原