基于多特征融合的Web图像聚类算法

Web Image Clustering Algorithm Based on Multi-features Fusion

提出了一种融合多种图像特征信息的Web图像聚类算法。本文的创新点主要表现在:将具有"图像标签","时间信息"",地理信息"以及"视觉特征"等多种特征的Web图像的聚类问题转换为K分图的划分问题。接下来将K分图的划分转化为若干个二分图的加权划分问题,利用对角矩阵和拉普拉斯矩阵,对K-1个目标函数进行线性加权,利用二次约束二次规划而完成对K分图的划分。通过对上述过程进行迭代运算得到Web图像的聚类结果。实验结果表明,本文算法通过对图像的多种特征信息的有效融合,降低了聚类的错误率,有效提高了Web图像聚类性能。

This paper proposes a Web image clustering algorithm by integrating a variety of image features. The main in- novations of this paper lie in that Web image clustering problem is converted to the K-partite graph partitioning problem, and the multi-features of Web images include ＂image tag＂, ＂time information＂, ＂geographic information＂, and ＂visual features＂. Afterwards, K-partite graph partitioning problem is solved by several weighted bipartite graphs partitioning. Based on the diagonal matrix and Laplacian matrix, K-1 linear weighted objective functions are calculated by secondary constrained quadratic programming to obtain the results of K-partite graph partitioning. Next, the Web image clustering results can be obtained by iterative calculation of the above process. The experimental results show that the proposed al- gorithm can effectively reduce the error rate of the clustering and then improve the performance of Web image clustering through the effective multi-features fusion