基于多尺度分解多源遥感图像的融合技术研究

Considering the characteristics of multi-source remote sensing image that are multiple perspectives, multiband, multi-scale and multi-period, a multi-period - air - spectrum image sequence integration fusion theory model and method are established. Through the registration parameters, outliers position combined the iteration and the initial registration parameters on multi-source image data from the group of point detection, realize registration parameters and outliers synchronous optimization. Using maximum likelihood method to the model parameters of outliers and fitting model to solve the standard deviation and outliers distribution ratio parameters to structure image registration and realize high precision registration with the objective function. Research on image fusion method of low frequency and high frequency component based on multi-scale decomposition, and establish the corresponding mathematical relationship between the observed image and the fusion image. In the Bayesian maximum a posteriori probability criterion, derivate an integrated image fusion theory framework and constructs the image space and spectrum function model to form the integration of the formation of stable image fusion model.On the condition of constructing spectrum constraint model and taking account of regional characteristics of the edge image space constraint model, to realize the high integration of spatial information and high fidelity of spectral information. Finally get a simple and effective fusion method of data driven.

针对多源遥感图像多角度、多波段、多尺度、多时相等特征，建立多时-空-谱图像序列的一体化融合理论模型与方法。通过对配准参数和离群点位置进行联合迭代求解，初始配准参数对多源图像数据进行离群点检测，实现配准参数和离群点同步优化；利用最大似然方法对离群点的模型参数进行拟合，求解模型的标准偏差和离群点分布比例参数，构造图像配准的目标函数，实现高精度配准；研究基于多尺度分解的图像融合方法中的低频和高频分量的融合方法，描述观测图像与融合图像之间对应的数学关系；在贝叶斯最大后验概率准则下，经过理论推导建立一体化图像融合理论框架，并构建图像的空间和光谱函数模型，形成稳健的一体化图像融合模型。在构建光谱约束模型及顾及区域特征的保边缘图像空间约束模型条件下，同时实现空间信息的高融入与光谱信息的高保真，最终得到一种数据驱动且简单有效的融合方法。

研究一种基于多尺度小波分解的图像融合方法。利用小波变换对每一幅图像进行多尺度分解，对各分解层上不同频带的子图像采用不同的融合处理，即对多尺度分解后的低频分量和高频分量分别融合，并把成分替换法与多分辨率分析法相结合，以实现多光谱遥感图像分辨率增强的全色锐化图像融合。采用面向对象的区域加边缘有效因子的融合思想实现分区融合，使得融合的图像最大限度地保留了多光谱图像的光谱信息和高的空间分辨率。