并行图像配准算法在幸运成像中的应用

Application of Parallel Image Registration Algorithm in Lucky Imaging

借助图形处理器(GPU)在通用计算领域的优势,解决图像配准面临的处理速度问题。研究了基于GPU加速处理图像配准的算法;根据Fourier-Mellin变换的图像配准算法原理,提出相应的GPU并行设计模型;利用计算统一设备架构的软硬件体系架构,实现Fourier-Mellin变换算法向GPU的移植。实验表明,运用所提出的并行方案完成分辨率1 024×1 024像素的图像配准耗时22ms,有效提升了图像配准效率,增强了幸运成像技术工程应用的可能性。

Taking advantage of the computing power of graphic processing unit （GPU） in general purpose computation, this paper studied the accelerated image registration algorithm based on GPU to solve the processing speed problem of image registration. Based on the image registration principle of Fourier-Mellin transform,a parallel registration algorithm model based on GPU was proposed. By employing the software and hardware architecture of compute unified device architeeture（CUDA）, the transplant of Fnurier-Mellin onto GPU was finally achieved. The experimental results show that it takes about 22 ms for the proposed parallel computing algorithm to register images with the resolution of 1024 × 1024 pixels, effectively enhancing the efficiency of image registration and the possibility of its application in lucky imaging engineering.