Sparse Multi-Modal Hashing.pdf

IEEE TRANSACTIONS ON MULTIMEDIA, VOL. 16, NO. 2, FEBRUARY 2014 427 Sparse Multi-Modal Hashing Fei Wu, Zhou Yu, Yi Yang, Siliang Tang, Yin Zhang, and Yueting Zhuang Abstract—Learning hash functions across heterogenous high-di- mensional features is very desirable for many applications in- volving multi-modal data objects. In this paper, we propose an approach to obtain the sparse codesets for the data objects across different modalities via joint multi-modal dictionary learning, which we call sparse multi-modal hashing (abbreviated as ). In , both intra-modality similarity and inter-modality sim- ilarity are first modeled by a hypergraph, then multi-modal dictionaries are jointly learned by Hypergraph Laplacian sparse coding. Based on the learned dictionaries, the sparse codeset of each data object is acquired and conducted for multi-modal approximate nearest neighbor retrieval using a sensitive Jaccard metric. The experimental results show that outperforms other methods in terms of mAP and Percentage on two real-world data sets. Index Terms—Dictionary learning, multi-modal hashing, sparse coding. I. INTRODUCTION S IMILARITY search, a.k.a., nearest neighbor (NN) search,is a fundamental problem and has enjoyed great success in many applications of data mining, database, and information retrieval. With the explosive growth of high-dimensional data, e.g., the images and videos on the web, there is an emerging need of the NN search on high-dimensional feature space. The problem of NN search can be described as follows: given a query data object , finding the top- nearest neighbors to the query from a target data set. The simplest way to solve the NN search problem is the brute- force linear search. However, this becomes prohibitively expen- sive when the number of retrieved target data objects are very large scale. To speed up the process of finding relevant data ob- jects to a query, indexing techniques are necessarily conducted to organize target data objects. However,