均值漂移算法的研究与应用.pdf

第22 卷第8 期 控 制 与 决 策 2007 年8 月 V ol . 22 No . 8 Control and Decision Aug . 2007 : 100 10920( 2007) 08084 107 1 1 2 周芳芳 , 樊晓平 , 叶 榛 ( 1. , 4 10075; 2. , 100084) : 对均值漂移算法的理论和应用作一全面的综述. 首先根据密度函 的非参 估计推导出均值漂移公式的一 般形式, 说明了均值漂移迭代算法的步骤及收敛性; 然后重点讨论核函 的选择以及带宽矩阵的计算等关键技术; 最 后归纳了均值漂移算法在模式检测聚类图像分割以及物体实时跟踪等方面的应用, 并展望了均值漂移算法在理论 和应用中的研究方向. : 均值漂移; 非参 密度估计; 核函 ; 带宽 : T P301 : A Mean shift research and applications 1 1 2 ZH OUFangf ang , FA N X iaop ing , YE Zhen ( 1. Schoo l of Info mation Science and Enginee ing , Cent al So ut h U niv e sity , Chang sha 4 10075, China; 2. State K ey L abo at o y of Intelligent T echnolog y and Sy stem, T sing hua U nive sity, Beij ing 100084, China. Co espo ndent : ZH O U Fangfang, Email : zff@ mail . csu. edu . cn) Abstract: A co mp ehensive su v ey of the uptodate esea ch and applicatio n fo mean shift a e p esented. A g ene al mean shift p ocedu e is def ined by no npa amet ic est imatio n of density g adient, and the steps and co nv e gency of the algo ithm a e p o po sed. T he key techniques of the selection of ke nel function and the computatio n of bandwidth mat ix a e int o duced em phatically . F inally, the applicat ions of mean shift to mo de detectio n, cluste ing, image segmentation and t acking a e su veyed in deta il and t he di ections tow a ds t he futu e esea ch a e o ut lined. Key words : M ean shift; N onpa amete density estimat ion; K e nel f unction; Bandw idth 1 引 言 , ( M S) , 1 , [ 1] , Fukunaga 1975 . 1995 , Cheng[ 2] MS . , , , .