基于爬山法的快速图像匹配  

Fast Image Matching Based on Hill-climbing

作　　者：闫宇壮[1] 王亦平[1] 黄新生[1] 

机构地区：[1]国防科技大学机电工程与自动化学院,长沙410073

出　　处：《科技导报》2008年第20期72-75,共4页Science & Technology Review

基　　金：兵器预研项目(2020203)

摘　　要：针对高速系统的应用需求,分析一种基于爬山法的快速图像匹配算法。文章首先通过分析NNPROD算法的相关度量曲面,设计了多个爬山者按照优先级依次爬山的启发式搜索算法,该算法能够保证全局最优。然后分析了进一步提高算法快速性的三种改进方案:爬山步长的自适应调整、通过图像样本集来计算相关度、爬山与分层相结合的搜索策略。实验结果表明,爬山搜索算法性能优越。目前,该法已在实际系统中得到应用。For a high speed system, a fast image matching algorithm based on Hill-climbing search is presented. By analyzing the correlative surface of NNPROD, a heuristic searching algorithm is defined for muhi-climbers to climb the hill according to their priorities to ensure the glohal optimization. The paper also proposes three schemes to improve the speed of image matching, they are： adaptive adjustment of the climber＇s step, calculation of the correlative coefficient according to samples of the images, and a strategy of combining the Hill-climbing and Layered-searching. The experimental results show that the algorithms presented enjoy an excellent performance.

关 键 词：爬山法 快速算法 分层搜索 图像匹配 NNPROD算法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]