Adaptive multi-modal fusion hashing via Hadamard matrix

Abstract

Hashing plays an important role in information retrieval, due to its low storage and high speed of processing. As an effective multi-modal representation learning method, multi-modal hashing has received particular attention. Most of the existing multi-modal hashing methods adopt the fixed weighting factors to fuse multiple modalities for any query data, which cannot capture the variation among different queries. Besides, there are too much hyper-parameters in their models while it is time-consuming and labor-intensive to determine the proper parameters. The limitations may significantly hinder their promotion in practical applications. In this paper, we propose a simple, yet effective method that is inspired by the Hadamard matrix. On the one hand, our proposed method that involves a very few hyper-parameters is flexible. On the other hand, the complementary information between multi-modal data and the semantic discrimination information are preserved well in the hash codes. Extensive experimental results on four benchmark datasets show that the proposed framework is effective and achieves superior performance compared to state-of-the-art methods.