摘要
为了提高基于内容的图像检索系统的检索速度,提出了一个基于快速聚类索引的图像检索算法,并将其应用于视频新闻检索系统中。该算法采用Fastmap算法实现图像高维特征向量降维,并用改进后的模糊C均值聚类算法对降维后的图像进行聚类,生成图像索引。该算法用于图像检索,检索时间不会随着图像数据库中图像数量、特征向量维数的增加而增加,极大地提高了系统的检索效率,有效地解决了聚类中心初试值的选取问题。同时利用该算法构成的系统还具有动态删除、分裂、合并、插入等功能。实验结果表明,与顺序扫描算法相比,该系统不仅大大提高了检索速度,而且在图像数目和特征向量空间维数增大的条件下,仍能够获得良好的检索性能。
In order to construct effective organization and index to enhance speed of retrieval,a fast clustering scheme for images indexing in content-based image database was proposed. The dimension of the features of images are reduced by Fastmap algorithm, and then for the new features images are clustered by a modified fuzzy c-means clustering algorithm so that images can be matched within the corresponding clusters. The system using the scheme as the searching engine has the merits as following: choosing of the centroids of clusters properly , image eliminating, assigning, splitting, unitting, merging and inserting dynamically with high efficiency.
出处
《吉林大学学报(信息科学版)》
CAS
2004年第6期638-642,共5页
Journal of Jilin University(Information Science Edition)