Binary Multi-view Image Re-ranking

Zhijian Wu, Jun Li, Jianhua Xu

Abstract: Conventional subspace-based multi-view re-ranking methods essentially handle the Euclidean feature space transformation and tend to be inefficient when dealing with large-scale data, since the cost of computing the similarity between the query item and the database item is prohibitively high. Inspired by Hashing technique, in this paper, we propose an efficient binary multi-view image re-ranking strategy in which the original multi-view features are projected onto a compact Hamming subspace. With the intrinsic structure of the original multi-view Euclidean feature space maintained, the resulting binary codes are consistent with the original multi-view features in similarity measure. Furthermore, coupled with the discriminative learning mechanism, our method leads to compact binary codes with sufficient discriminating power for accurate image re-ranking. Experiments on public benchmarks reveal that our method achieves competitive retrieval performance comparable to the state-of-the-art and enjoys excellent scalability in large-scale scenario.