基于信息粒度的聚类分析及其应用

Analysis and Application of Clustering Based on Information Granularity

在处理复杂问题时,通过改变问题所在的粒度空间,不仅可以有效获取对象的特征,而且可去除干扰和非本质属性,使问题易于分析解决。所谓从粒度计算的观点来讨论聚类分析问题,就是认为聚类是在原问题的粒度下(同一问题的最细粒度空间)进行问题分析。为了简化处理,引入不同的聚类相似性函数,其实质就是得到不同粒度空间的等价类。在实际问题求解中,可以根据问题需要取不同相似性函数,以便将问题变换到所需的粒度空间进行处理。为推广其应用,将该思想应用于车牌二值化,提出了基于信息粒度的聚类变换的二值化算法,实现了从彩色3维空间到黑白1维空间的粒度变换。实验结果表明,该算法所得结果更加切合实际图像,不仅具有普适性,而且有利于下一步的识别操作,尤其对于各种斜车牌、光照不均车牌更具有一定的优越性。

In dealing with complicated problems, the characters of the object can be obtained effectively when the disturbing and nonessential attribute can be wiped off by changing the granular space where the problem located, which make it easier to analyze and solve the problems. In this paper, the analysis of clustering is discussed according to granularity computing. It is assumed that the clustering problems are analyzed under the same granularity （the finest granular space of the problem）. The essential of introducing the different comparability functions of clustering is to get a series of equivalence species of different granular space. In practice, problems can be transformed into required granular space, by selecting different comparability functions according to the problem. The transformation form multicolor three- dimension space to monochrome one-dimension can be realized by proposing The License Plate Binary Algorithm based on Information Granularity. Experiments show that the results of this algorithm are more suitable to actual image, have broad generality, and are in favor of recognition following. It is especially predominant in inclined plates or asymmetrical illumination plates.