图像统计模型参数估计中的期望最大值算法

Expectation maximization method for parameter estimation of image statistical model

期望最大值算法是近年来图像统计模型参数估计技术领域的研究热点之一。在对期望最大值算法分析的基础上,结合其在图像统计模型参数估计中的应用研究,对改变标准期望最大值算法的3种方式进行比较分析。结合图像恢复、分割、目标跟踪以及与其他优化算法的融合应用,从丢失数据集的选取、丢失数据集和不完全数据集统计模型的建立,以及统计模型参数估计3个方面,评述期望最大值算法优缺点。丢失数据的选取和不完全数据的描述形式直接决定期望最大值算法的结构和计算复杂度,以致算法的成败。最后,讨论期望最大值算法目前存在的问题及未来的发展方向,指出其在具有丢失数据统计模型参数估计中广泛应用。

Expectation maximization （EM） algorithm for parameter estimation of image statistical model is one of the striking research fields in recent decades. Based on the analysis of the EM algorithm, combining the current application research in parameter estimation of image statistical model, analysis and comparison are conducted in terms of the three improvement schemes of standard EM algorithm. In this paper, integrating image restoration, segmentation, object tracking and the fusion of other evolution optimization algorithms, through three aspects, such as the selection of missing data sets, the statistical model establishments of missing and incomplete data sets, and parameter estimation of image statistical models, as well as the advantages and disadvantages of the corresponding EM algorithm are exponded. The structure and complexity of EM algorithm, so far as to success or failure, are directly determined by the selection of missing data and the expression form of incomplete data. In the end, challenges and possible trends are discussed, and extensive applications of EM algorithm to parameter estimation of statistical model with missing data are pointed out.