Mean Shift自适应步长的改进  

Improvement of Adaptive of Mean Shift Algorithm

作　　者：黄家祥[1] 李绍滋[1] 成运[2] 张帆[3] 

机构地区：[1]厦门大学信息科学与技术学院,福建厦门361005 [2]湖南人文科技学院通信与控制工程系,湖南娄底417000 [3]华北水利水电学院信息工程学院,河南郑州450011

出　　处：《厦门大学学报（自然科学版）》2009年第2期189-193,共5页Journal of Xiamen University：Natural Science

基　　金：国家自然科学基金(60873179)资助

摘　　要：Mean Shift算法是一种广泛应用于计算机视觉、模式识别等领域的统计迭代算法,它使每一个点"漂移"到密度函数的局部极大值点,均值漂移的方向就是梯度方向,因此,漂移序列总是向着函数值增加最快的方向移动,并且每次移动的步长大小具有自适应性.本文研究了Mean Shift算法移动步长的自适应性,对其进行改进,使其能够通过参数的适当调整得到优于原Mean Shift算法的收敛速度,并从理论上证明了改进的Mean Shift算法能够收敛.本文的实验也进一步验证了改进的Mean Shift算法的收敛性,并对比了改进前后的Mean Shift算法的收敛速度.Mean Shift algorithm is a statistics iterative algorithm which is widely used in computer vision and pattern recognition, it makes each point ＂shift＂ to the local maximum of density function. Shifting direction is gradient direction of density function at each point,accordingly,shift point is always moving towards the direction which makes density function increase fastest, furthermore,the move length is adaptive. This paper investigates the adaptive of move length of each point in Mean Shift algorithm and improves it, enhances its convergenee rate, moreover, proves the convergence of the improved Mean Shift algorithm. Experiment in this paper validates the convergence of the improved Mean Shift algorithm,contrast convergence rate of improved Mean Shift algorithm with convergence rate of traditional Mean Shift algorithm.

关 键 词：Mean SHIFT算法 自适应性 收敛性 收敛速度 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]