摘要
传统的模板图像匹配算法,匹配速度较慢。应用GPU通用高性能编程技术实现了一种加速图像匹配算法的新方法。应用CUDA编程技术对图像匹配算法进行并行化改造。采用了四种不同的存储方案,在第四种存储方案中获得了43.5倍的加速比,并对四种不同的存储方案的性能进行了深入研究。
The speed of conventional image matching is very slow.A novel approach is presented which accelerates the image matching algorithm by using the GPGPU technology,and CUDA is used to reprogram the process.Four methods by using several types of cuda device memory are proposed.In the fourth method,43.5 times speedup has been achieved in the GPU as compared to the CPU counterpart,which demonstrates the significance of parallelizing the image matching algorithm using the GPGPU technology in image matching system.The importance of memory access efficient as a limiting factor to parallelism is investigated.
出处
《计算机工程与应用》
CSCD
2012年第2期173-176,共4页
Computer Engineering and Applications
关键词
图像匹配
存储模型
图形处理器(GPU)
并行计算
image matching
Compute Unified Device Architecture（CUDA） device memory types
Graphic Processing Unit（GPU）
parallel computing
