动态的K-均值聚类算法在图像检索中的应用  

Dynamic k-mean clustering algorithm applied in image searching

作　　者：张白妮[1] 骆嘉伟[1] 汤德佑[1] 

机构地区：[1]湖南大学计算机与通信学院,湖南长沙410082

出　　处：《计算机工程与设计》2004年第10期1843-1846,共4页Computer Engineering and Design

摘　　要：聚类分析技术已经广泛应用于基于内容的图像信息挖掘领域,该技术提高了图像检索的速度和质量。K-均值算法和自适应算法是两个典型的聚类分析算法,但K-均值算法严重依赖于经验参数和阙值的设定;自适应算法得到的聚类个数太多,相应的就是类内的图像个数过少,效率不是很高。从选取初始聚类点是否具有确定性、迭代次数是否过多和聚类个数是否适当等方面考虑,提出了一种新的聚类算法,即动态的K-均值法。模拟实验的结果表明,该算法具有较好的准确性和效率,使检索的质量和速度都得到了很大的提高。Clustering and analysizing techniques has already been abroad applied in content-based image information mining fields. This technique improves the speed and the quality of image searching. K-mean clustering algorithm and self-adapt algorithm are two typical clustering and analyzing algorithms. But K-mean clustering algorithm heavily depends on the setting of experienced parameters and hold value. Self-adapt algorithm gets too many numbers of classes that the numbers of images only are few in each class, so the eddiciency is low. Being considered whether choosing the clustering sets make certain, whether iteration is too much and whether the number of clustering class is proper, a algorithm named dynamic k-mean clustering algorithm is put forward. Simulation experiment shows that this algorithm is of better veracity and efficiency, so that the quality and the speed of searching are advanced greatly.

关 键 词：K-均值聚类 图像检索 K-均值算法 基于内容 聚类算法 自适应算法 图像信息 个数 速度 技术 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] G633[自动化与计算机技术—计算机科学与技术]