摘要
基于图像处理器(GPU)平台提出了维纳并行滤波算法,解决了航空成像的像移实时补偿问题。介绍了原有像移补偿算法及其PC运行平台的不足,立足于现有图像处理器通用计算技术对原有像移补偿算法做多线程并行计算的改进。针对图像处理器件的硬件并行架构,优化设置算法多线程访问,提高了各个线程的访问速度。该环节的改进甚至能将算法效率提升到原来的3倍。借助图像处理器的并行运算优势,单帧1 024×1 024灰度图像的恢复处理时间只需要8 ms,整个算法的运行速度达到原先的20倍,能够完全满足高分辨率航拍视频图像的像移模糊实时恢复的需求。
In order to improve the speed of restoration algorithm, a parallel Wiener filtering method based on Graphic Processing Unit(GPU)platform is presented to restore motion-blurred aerial image degraded in a deterministic way by motion or vibration. The shortages of the original algorithm using Wiener filtering and original PC run-time platform are introduced. On the basis of the new General Purpose GPU (GPGPU) technology, the original algorithm is divided into thousands of single algorithm threads to be computed in parallel. According to the special simultaneous operating mode of GPU hardware, a way in which the algorithm threads access the data on the GPU global memory is specially configured to improve the accessing speed, the algorithm efficiency by the special configuration can even be improved roughly 3 times that by original configuration. With the parallel computing ability of GPU, the new algorithm can restore 1 024×1 024 gray image in 8 ms per frame. The experimental result shows the new algorithm based on GPU reaches approximately 20 times that of original algorithm based on CPU of personal PC, which can completely be applied to the real time restoration of high resolution motion-blurred aerial image.
出处
《光学精密工程》
EI
CAS
CSCD
北大核心
2009年第1期225-230,共6页
Optics and Precision Engineering
基金
国防科技预研基金资助(No.1040603)
关键词
航空成像
并行计算
维纳滤波
图像恢复
图像处理器
aerial imaging
parallel computing
Wiener filtering
image restoration
Graphic processing Unit (GPU)