Diffusion Hashing

摘要：

With the worldwide spread of the broadband Internet, massive multimedia data including texts, images, and videos are increasing explosively and available for interactive applications over the Internet. At the same time, more and more attention has been paid to aiming at fast retrieval from massive multimedia databases. Hash-based Approximate Nearest Neighbor (ANN) search is a technology that achieves fast retrieval by regarding the hash key as a retrieval index, where the similarity of data is maintained and embedded in the neighborhood of the hash key. In other words, the closer the Hamming codes between hash keys, the more similar the data become. In general, short binary codes are preferred for storing hash keys and values. The difficulty is to define the similarity between data and reflect it in binary codes. In this paper, we propose Diffusion Hashing (DH) as a novel ANN search technique based on hashing with an anisotropic diffusion kernel. DH aims to transform the search index into as short binary codes as possible, preserving the similarity induced by random walk on the data manifold in higher dimensional space. From comparative experiments, we will demonstrate that DH outperforms previously known hash-based ANN search techniques including Locality Sensitive Hashing and Spectral Hashing.

作者: Atsushi Tatsuma Masaki Aono

作者单位: Toyohashi University of Technology, Aichi, Japan

会议类型: 国际会议

会议名称: 2011亚太信号与信息处理协会年度峰会(APSIPAASC 2011)

会议地点: 西安

会议语种:英文

页码: 1-7

在线出版日期: 2011-10-18（万方平台首次上网日期，不代表论文的发表时间）