基于多核CPU的太阳自适应图像斑点重建技术  

Adaptive Solar Speckle Image Reconstruction Based on Multi-Core CPU

作　　者：唐若林 田雨[1,2] 钟立波 饶长辉[1,2] 

机构地区：[1]中国科学院自适应光学重点实验室,四川成都610209 [2]中国科学院光电技术研究所,四川成都610209 [3]中国科学院大学,北京100049

出　　处：《激光与光电子学进展》2017年第6期98-103,共6页Laser & Optoelectronics Progress

基　　金：中国科学院国防科技创新基金项目(YJ16K006)

摘　　要：地基太阳高分辨力成像中,自适应光学系统补偿了图像中心等晕区的像差,需要斑点图像重建技术用来产生全视场衍射极限分辨率的图像。一个从Matlab程序移植而来的C语言程序被设计用于加速处理。该程序采用斑点干涉法重建图像的傅里叶振幅和斑点掩膜法重建图像的傅里叶相位。使用OpenMP进行加速,使核心间共享了部分内存资源,图像按子块进行并行计算。移植过程中程序算法上进行了必要的优化并移除了大量的冗余计算。程序使用英特尔ICC编译器编译,运行在一个12核的Linux服务器上。一张1280pixel×1280pixel的图像可以在31s内重建完成。相对于单核运行,加速比最高可以达到10.66。单台服务器上相对于并行接口获得更好的扩展性。In solar ground-based high resolution imaging, the speckle image reconstruction is used to yield diffraction-limited resolution for partially corrected images with the solar adaptive optics system. The C language program which adopts speckle interferometry to reconstruct the Fourier amplitudes and speckle masking to reconstruct the Fourier phase is generated from Matlah procedure. The speckle image reconstruction is based on the multi-core central processing unit architecture. The C program is parallelized by OpenMP, and the program is partitioned by sub-image. In addition, the redundancy calculating is removed. This program is compiled by Intel ICC and run on a 12 cores server with Centos 7.0. As a result, a 1280 pixel× 1280 pixel object image can be reconstructed from 100 speckle images within 31 s, Compared with the program run on single core, the speedup is around 10.66. The scalability is better than the program that parallelized by message-passing interface is single server.

关 键 词：图像处理 图像重建技术 斑点干涉法 斑点掩膜法 并行计算 

分 类 号：TN911.73[电子电信—通信与信息系统]