中文摘要：

目的：Mean shift方法在模式检测、聚类、图像分割、图像滤波以及目标实时跟踪等方面的应用非常广泛。本文主要对Mean shift基本方法进行理论分析并对其在图像处理与模式识别领域的应用进行综述和展望。方法：首先根据非参数密度估计理论推导Mean shift方法的一般公式与表达形式,并对算法的基本步骤和收敛性进行了分析和论述;然后对核函数的选择以及窗口带宽矩阵的计算等关键技术进行讨论。结果：在综述mean shift基本原理基础上对该方法的特点、发展及应用进行了展望。结论：Mean shift算法具有计算简单、收敛速度快和对噪声的鲁棒性强等优点。作为一种高效的非参统计迭代算法,Mean shift方法在很多领域都得到了广泛的应用,未来还会在图像处理和模式识别领域得到更大的发展及应用。由于该算法的计算复杂度严重依赖于Mean shift公式中核函数的选择和带宽矩阵的计算,该方法对于多变量、多模态数据的处理速度和效果尤其有待提高。因此,该算法未来的主要研究方向为对多模态和多变量数据的快速处理方法研究。

英文摘要：

Objective：Mean shift method has been widely used in fields of mode detection,clustering,image segmentation and real-time tracking.In this paper,a comprehensive survey and outlook for Mean shift approach and its applications in image processing and pattern recognition is presented.Methods：Firstly,a general formula of Mean shift procedure is defined by deduction based on theory of nonparametric estimation of density gradient,and the steps and convergency of the method are also proposed.Afterwards,the key techniques such as selection of kernel function and the computation of bandwidth matrix are discussed.Results：Based on comprehensive investigation about fundamental principles of Mean shift method,the futurere searches about its developments and applications are outlined.Conclusions：Mean shift method has such merits as simplecomputation with high converging speed and robust stability at presence of noise.As a nonparametric and recursive statistic algorithm with high efficiency,Mean shift is widely used in various fields and will be extent to even wider use in field of image processing and pattern recognition.For complexity of this algorithm depends seriously on selection of kernel function and computation of bandwidth matrix in Mean shift formula,the processing speed together with the final results is needed to be improved especially in multi-variable and multi-mode occasions.Therefore,the future research should emphasize on fastapproach for multi-variable and multi-mode processing.