基于聚类的图像拷贝检测技术仿真研究

Simulation Research on Clustering-based Image Copy Detection Technique

在研究进行图像拷贝检测检索技术时,传统的图像拷贝检测检索使用的是顺序检索。针对目前大容量、高维度的图像数据集,传统检索方法效率低,无法有效发现非法拷贝,阻止盗版行为。为提高检索效率,对图像数据库进行必要的预处理并建立高效索引,首先利用k-d树构建图像多维特征的索引结构,对其进行改进,使之适应于维度较高的情况,再结合聚类处理思想,对k-m eans聚类方法进行改进,实现对高维特征向量做聚类预处理,最后再基于聚类中心做k-d树的索引结构。实验结果表明,上述方法在满足大规模图像拷贝检测的效率的同时,其查全率和查准率都有较大提升,为打击盗版提供技术支持。

The traditional image copy detection methods are inefficient and cannot meet the practical requirement in the case of large-volume and high-dimension image datasets.This paper attempts to improve k-means clustering method,and process the multidimensional eigenvectors.Then a k-d tree index structure is built based on the clustering centre.The experimental results show that this method can improve both precision and recall ratios while meeting the retrieval efficiency in the case of large-volume and high-dimension image data.