多尺度变换域隐马尔可夫模型能够有效地描述变换域系数在尺度间、尺度内和方向间的统计相关性,是

一种新的统计图像感知与识别方法. 文中以变换域系数的统计相关性描述为中心,以模型的设计和应用的开展为

两翼,深入分析了子波变换的三级统计特性与机理,比较研究了多尺度变换域的十种统计模型,并系统评述了这些

模型在图像感知、处理和分析中的最新进展. 同时,具体论述了这一领域研究中两类成功的实例:图像去噪和图像

纹理分割. 对于前者,以Lena 图像为测试用例分析比较了以变换域统计模型为核心的8 种算法的去噪性能 对于

后者,按照分割类型(监督式或非监督式) 和应用的图像类型系统比较了以统计模型为基础建立的15 种图像分割

方法. 最后,从面向应用的模型构造和算法设计、变换域的拓展和应用层次的推广三个层面指出了目前存在的问题

和不足,探讨了进一步的研究重点-　Multiscale t ransform2domain hidden Markov model s ( HMM’s) , a class of new statisti2

cal approaches to image perception and recognition , can effectively characterize the inter scale , in2

t rascale and cross2orientation correlations of t he coefficient s in different multiscale t ransform do2

mains. Wit h t he statistical characterization of t he coefficient s in t ransform domains t he center and

t he design of various statistical model s and t he development of a variety of applications two

wings , t hree classes of statistical characteristics and their mechanisms of discrete wavelet t rans2

form of an image are analyzed in dept h , ten statistical model s in multiscale t ransform domains are

st udied comparatively , and t he state2of 2t he2art of multiscale t ransform2domain statistical model s

in image perception , processing and analysis is systematically reviewed. Two types of successf ul

applications in this research domain are discussed in detail , i. e. , image de