基于小波的遥感图像全局配准算法研究及其并行实现

An Automatic Global Registration Algorithm Based on Wavelet and Its Parallel Implementation

随着遥感技术的发展，遥感图像处理领域对自动图像配准技术的需求越来越迫切．首先提出了点匹配和全局配准相结合的自动图像配准算法；从实现快速配准的途径出发，提出了利用多分辨率小波缩小搜索空间及进行全局配准的自动算法，详细设计了逐级求精的搜索策略，并比较总结了算法的特点；在此基础上，提出了两种可行的数据并行方案；最后在一个小规模的机群系统上实现了上述串、并行算法，给出了客观的性能评价．实验结果表明文中提出的算法达到了预期的目标，即针对多传感器、大数据量的遥感图像，在保证精度的前提下，进行了快速高效的自动配准．

Image registration is widely used in image processing.With the developmentof remote sensing technology,the demand of fast and automatic image registration becomesmore urgent.Based on a thorough investigation on automatic registration methods for re-mote sensing image,a new automatic image registration algorithm is proposed by combiningthe point-to-point matching with the global registration.The focus is on fast registration,whereby a wavelet-based automatic global registration algorithm is designed,which relieson iteratively refining strategy to reduce the search space.The features of the algorithm areconcluded after comparison with some other methods.In addition,two parallel computationschemes are proposed and implemented on a cluster system with 16 processors.The experi-ment results show that the proposed algorithms could achieve high computational efficiencywith satisfactory registration accuracies for large remote sensing images.