Determining appropriate master images, reducing radiometric error accumulation, and eliminating outliers from the cloud, water, and land changes, are three main issues in radiometric normalization of multitemporal high-resolution satellite images (HRSI) during mosaicking. However, these three issues have not been simultaneously considered by the existing methods. This article presents a comprehensive radiometric normalization method for multitemporal HRSI using a radiometric block adjustment without master images. Pseudoinvariant features (PIFs) extracted from image pairs using the iteratively reweighted multivariate alteration detection are used as the corresponding pixel observations and organized to form radiometric tie points according to the corresponding horizontal space coordinates. Radiometric error equations are subsequently constructed, and the linear radiometric transformation parameters are solved by a global adjustment. The time-invariant PIFs generally represent the true corresponding features and naturally avoid the cloud, water, and land changes, which can eliminate the effects of outliers. Furthermore, the pixel values of tie points calculated from the weighted average of the corresponding pixel observations are used as virtual radiometric control points to eliminate the dependency on master images. Moreover, a global optimum can be achieved by the global adjustment, effectively overcoming the error accumulation, which is severe in large datasets. Four groups of HRSI datasets from various satellites are used to validate the performance of the proposed method. Experimental results demonstrate that the proposed method outperforms two state-of-the-art methods and has good applicability and stability, considering both visual effects and quantitative performance.

中文翻译：

---

通过块平差实现多时相卫星图像的鲁棒辐射归一化，无需主图像


确定合适的主图像、减少辐射误差累积以及消除云、水和土地变化引起的异常值是镶嵌过程中多时相高分辨率卫星图像（HRSI）辐射归一化的三个主要问题。然而，现有方法并未同时考虑这三个问题。本文提出了一种使用无需主图像的辐射块平差的多时相 HRSI 综合辐射归一化方法。使用迭代重加权多元变更检测从图像对中提取的伪不变特征（PIF）用作相应的像素观测值，并根据相应的水平空间坐标组织形成辐射连接点。随后构建辐射误差方程，并通过全局平差求解线性辐射变换参数。时不变的PIF一般代表了真实的对应特征，自然避免了云、水、地的变化，可以消除异常值的影响。此外，根据相应像素观测值的加权平均值计算出的连接点的像素值被用作虚拟辐射控制点，以消除对主图像的依赖。此外，通过全局调整可以实现全局最优，有效克服了大数据集中严重的误差累积问题。来自不同卫星的四组 HRSI 数据集用于验证所提出方法的性能。 实验结果表明，考虑到视觉效果和定量性能，所提出的方法优于两种最先进的方法，并且具有良好的适用性和稳定性。