基于一种改进禁忌搜索算法优化离散隐马尔可夫模型  

Optimization of Discrete Hidden Markov Model Based on an Improved Tabu Search Algorithm

作　　者：刘江华[1] 陈佳品[1] 程君实[1] 

机构地区：[1]上海交通大学信息存储研究中心,上海200030

出　　处：《计算机工程与应用》2003年第20期92-94,共3页Computer Engineering and Applications

摘　　要：隐马尔可夫模型(HMM,HiddenMarkovModel)是语音识别和手势识别中广泛使用的统计模式识别方法。文章提出了一种改进的禁忌搜索(ITS,ImprovedTabuSearch)优化HMM的参数。传统的TabuSearch(TS)与局部搜索算法(极大似然法)交替进行,从而加快了算法的收敛速度,并得到优化解。分别用TS及ITS训练隐马尔可夫模型进行动态手势识别。结果表明ITS可获得更高的识别率,且能达到全局优化。Hidden Markov model(HMM)is a statistical pattern recognition method that is extensively used in speech and gesture recognition.In this paper,an improved tabu search algorithm(ITS)is used to optimize the parameters of HMM.Traditional tabu search algorithm(TS)is combined with local search algorithm,maximum likelihood(ML )algorithm,which speeds the convergence of the algorithm and can get better solutions.TS and ITS are applied to training HMM for dynamic gesture recognition.The results show that ITS can get higher recognition rate,and find the optimal model parameters.

关 键 词：改进的禁忌搜索 隐马尔可夫模型 动态手势识别 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]