Mobile version of Hanspub

文章引用说明 更多>> (返回到该文章)

Shi, Q., Petterson, J., Dror, G., Langford, J., Smola, A.J. and Vishwanathan, S.V.N. (2009) Hash Kernels for Structured Data. Journal of Machine Learning Research, 10, 2615-2637.


  • 标题: 集成式位置敏感聚类方法Image Cluster Method Based on Ensemble Locality Sensitive Clustering

    作者: 彭天强, 高毫林

    关键字: 精确欧式空间位置敏感哈希, 随机映射, 图像聚类, 聚类集成Exact Euclidean Locality Sensitive Hashing, Random Projection, Image Clustering, Cluster Ensemble

    期刊名称: 《Artificial Intelligence and Robotics Research》, Vol.5 No.2, 2016-05-18

    摘要: 针对常用图像聚类尤其是图像视觉聚类方法聚类速度慢、不支持增量聚类的局限,提出了集成式位置敏感聚类方法。该方法首先根据聚类有效性指标估计合适的聚类数目,然后生成多重哈希函数,并用它们对各数据点进行映射得出多重桶标记,再对数据集各桶标记进行聚类得出多个基划分,最后对多个基划分进行集成得出最终划分。实验结果表明,在准确率方面,集成式位置敏感聚类在人工数据上与k-means结合聚类集成的方法相当,在图像集上与k-means结合聚类集成的方法接近。但集成式位置敏感聚类的优点在于其聚类时间快、适合于增量聚类等。因此,集成式位置敏感聚类方法可以用于解决高维图像特征聚类问题。 To overcome the weakness of k-means in image clustering especially visual image clustering, we proposed an Ensemble Locality Sensitive Clustering method. It first determined the number of clusters of dataset, then generated the multiple clustering resolutions based on Exact Euclidean Locality Sensitive Hashing algorithm, at last, cluster ensemble methods were applied to get final partition. The experiments on synthetic dataset and image dataset show that new method reaches the same level with k-means combined with cluster ensemble about clustering accuracy on synthetic data set, and slightly less accuracy on image dataset. But the advantage of new method is its clustering time is faster than k-means, and it is suitable for incremental clustering. Therefore, Ensemble Locality Sensitive Clustering is a promising clustering method for high dimension image data.