摘要
为了提高图模型方法的分割速度,本文提出该方法的一种并行实现方案.该方案通过网格划分来实现相似度矩阵的并行计算.同时考虑到相似度矩阵的稀疏性和矩阵向量乘运算的内在并行性,在该方案中本文设计并行 Lanczos 算法来求解特征值问题.在 MPI 环境下的实验结果表明,该并行方案是提高图模型分割方法实时性的有效途径.
A parallel solution of the graph-based method is proposed to improve the segmentation speed. In this solution, the similarity computation is parallelized by means of grid partition. And a parallel Lanczos algorithm is designed to compute the eigenvalues in view of the sparseness of the similarity matrix and the inner parallelism of matrix-vector multiplication. The experimental results under MPI environment show that the parallel solution effectively improves the real-time performance of the graph-based segmentation method.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
2007年第4期571-576,共6页
Pattern Recognition and Artificial Intelligence
基金
国家自然科学基金(No.60473014)