摘要
近似最近点搜索在很多领域是非常有用的算法,传统上主要实现方法是基于各种搜索树(K-D树)及变种。考察一种基于空间填充曲线的新型算法,使用基于任务的并行编程模型TBB库对其进行优化,实验表明该算法在低维度(D≤5)上有很好的性能和精确度。在大数据集上随着K值的增加,与处理器核心数目p的增加,性能和精确度均呈现出较强的伸缩性。
Approximate nearest neighbour search algorithm is a very useful algorithm in a lot of fields,traditionally its main realisation approach is based on various search trees(K-D tree) and their variations.In this paper a new algorithm based on space fill curve is reviewed, and the task-based parallel programming model with Threading Building Blocks library is utilised to optimise it.Experiment indicates that,in low dimension(D = 5) the algorithm has both the good performance and high precision.For large data set,along with the increase of K value and the number of process core p,both its performance and precision manifest quite strong scalability.
出处
《计算机应用与软件》
CSCD
2010年第10期266-268,278,共4页
Computer Applications and Software
关键词
最近点搜索
近似算法
空间填充曲线
并行处理
多核处理器
TBB
Nearest neighbour search
Approximate algorithm
Space fill curve
Parallel computing
Multi-core processor
TBB