作　　者： ; ; ;

机构地区： 深圳大学计算机与软件学院

出　　处： 《小型微型计算机系统》 2014年第1期114-120,共7页

摘　　要： 针对标志点一致性图像配准算法收敛时间较长问题,提出在CPU+GPU异构平台上的并行实现方法。该方法在GPU上完成了一致性图像配准算法中的弹性变换求解及逆函数求解的并行实现,在CPU上实现变量初始化以及迭代更新等过程,充分利用异构平台优势,提高该算法的运行效率。实验结果表明该并行实现方法可以获得较高的加速比. In view of the longer convergence time of consistent landmark-based image registration algorithm, a parallel implementation method is presented on the CPU ＋ GPU heterogeneous platform in this paper. The method has accomplished the parallel implementation of elastic transformation solution and inverse function solution of consistent landmark-based image registration algorithm in GPU, at the same time, it has fulfilled the processes of variable initialization and iterative update in CPU, making full use of the advantages of heterogeneous platform to improve the efficiency of the algorithm. The experimental results show that the higher speedup can be obtained in the parallel implementation method.

关 键 词： 加速比 图像配准 并行算法

领　　域： [自动化与计算机技术] [自动化与计算机技术]