基于SIFT特征点改进聚类的图像检索方法研究  

A Novel Image Retrieval Method Based on Combinatorial Improvement Clustering of SIFT Feature Points

作　　者：张磊磊[1] 陈丽芳[1] 

机构地区：[1]江南大学数字媒体学院,江苏无锡214122

出　　处：《软件导刊》2017年第9期188-191,共4页Software Guide

摘　　要：由于SIFT特征点能对图像局部特征进行合理、精确描述,有效使用SIFT特征点实现基于内容的图像检索成为当前计算机视觉领域中的热点问题。针对该问题,提出一种基于SIFT特征点的改进聚类的图像检索新方法。该方法包括图像颜色转换、特征点改进聚类算法,以及基于该算法的更有效的灰度直方图构建方法。与现有基于流光法的检索方法相比,该方法能有效解决聚类后特征点分组不确定和依赖特征点颜色信息和空间信息权重的问题。从公共图像库上的实验结果可以看出,该方法与现有方法相比具有较高的检索精度。Because SIFT feature points can represent the local feature of images effectively and accurately, how to realize content- based image retrieval by SIFT feature points is a hot issue in the computer vision. So, a novel image retrieval method based on the improvement combinatorial clustering of SIFT feature points is proposed. The proposed method consists of image color space transformation, improvement combinatorial clustering algorithm of SIFT feature points and effective local color histogram build- ing approach. Compared with existing methods based on FLOW feature points and local color histogram, the proposed method can solve the problem that uncertain sets of feature points after clustering and existing methods are sensitive to the weight of po sition and the color of feature points. According to experimental results on public image database, it can be seen that the pro posed method is more effective than existing methods.

关 键 词：图像检索 SIFT特征点 聚类算法 

分 类 号：TP317.4[自动化与计算机技术—计算机软件与理论]